MgmtWiki - User contributions [en]

Edge UX Engine

2026-03-26T01:11:19Z

Tom: /* References */

==Full Title==
Creating an edge user experience engine involves several key steps to ensure a seamless and efficient user experience. Here are some best practices:

==Context==

Typical process
# Conduct thorough research to understand the needs and preferences of your target users. This can involve surveys, interviews, and usability testing.
# Ensure that your edge engine is optimized for performance. This includes minimizing latency, optimizing resource usage, and ensuring smooth operation even under heavy load.
# Use a microservices architecture to build your edge engine. This allows for modular development, easier maintenance, and scalability.
# Security is crucial for any edge application. Implement strong authentication, encryption, and other security measures to protect user data and ensure privacy.
# Design your user interface with modern UI/UX principles in mind. This includes intuitive navigation, responsive design, and accessibility considerations.
# Regularly update and improve your edge engine based on user feedback and performance metrics. This ensures that your engine remains relevant and effective.

==Solutions==
* Edge Computing Platforms:
# [https://www.kwaai.ai Kwaai, pAI-OS] We are a volunteer based, open source AI research and development lab focused on democratizing artificial intelligence by building a Personal AI. We center our efforts around the principles of personal control, self-sovereign identity, transparency and openness.
# [https://www.getprimary.com/ The Workspace for Business] The Primary Browser is a fork of chromium with policy and IAM services.
# Install services like an IAM ([[Identity and Access Management]]) for others to access
# [https://mimik.com/ mimik edgeEngine] to develop and deploy your edge applications. These platforms provide the necessary tools and infrastructure to build efficient edge solutions. The edgeEngine enables any smart device to act as a cloud server with an API Gateway and runtime environment to host microservices. It is available for almost all smart devices: smart phones, PCs and tablets, IoT and embedded devices, PCBs, microcontrollers, etc.
# [https://www.linkedin.com/pulse/driving-innovation-collaboration-role-paios-shaping-future-ai-z3edc/ The Role of pAIOS in Shaping the Future of AI and Blockchain]

* API Gateways: Implement API gateways to manage communication between microservices and ensure efficient data flow.
* Containerization: Use containerization technologies like Docker to deploy and manage your microservices. This ensures consistency and scalability.
* Microsoft's Phoenix Project: Learn about Microsoft's efforts to modernize the Edge browser with UI upgrades and other enhancements here.

==Research and Testing==
True summative testing requires the messy reality of field experience.
Learning does not occur in sterile or isolated environments, and meaningful competence is often demonstrated only in complex, unpredictable contexts. Field-based experiences—such as internships, clinical practice, project-based work, or performance in authentic settings—capture skills like judgment, adaptability, collaboration, and problem-solving that controlled tests cannot fully measure. Overly controlled summative assessments risk privileging test-taking ability over real-world performance, producing results that are technically reliable but educationally shallow.
==References==

[[Category: User Experience]]

Artificial Intelligence

2026-03-23T00:46:53Z

Tom: /* Creativity */

==Full Title or Meme==
Intelligence without human lived experience or human compassion.

"The future is already here — it's just not very evenly distributed" - William Gibson.

The term [[Artificial Intelligence]] was coined, in part, out of spite. In 1955, the Dartmouth professor John McCarthy assembled a proposal for what became the first-ever conference on AI. McCarthy refused to use the prevailing term, "cybernetics": That would have necessitated inviting Norbert Wiener, then the world's leading cyberneticist, whom he found insufferable. Instead, McCarthy minted a new term that took off in the digital era, taking him along as a founder of the field. <ref>''Fun fact: Cybernetics'' Technology Review (2025-08) p 11</ref>

==Context==
===The Broad concepts===
* There are two great forces both in physics and in human history
** The harnessing of energy which started with the use of animal power to human benefit.
** The harnessing of information which started with philosophy and then Gutenberg's press.
* Likewise there are two great revolutions in human society.
** The Industrial revolution focused on control and use of energy which peaked in the 20th century.
** The Information revolution focused on control and use of information which is peaking in the 21st century.
* The social changes to accompany these revolution had to before the revolutions could complete.
** The educational system needed to be extended to all who work to engineer the control of energy.
** Now we need a second shift to educate most of the workers to control the information needed.
* The use of energy and information in harmony with the reality of the Earth's resources is a challenge.
** The profligate use of energy has altered the environment in ways that threaten human existence.
** The profligate use of information may prove to alter the environment of work to be a bigger threat.
* The struggle between conservative and progress forces in politics has been with us for ever.
** The conservative mind wants to hold onto the privilege of the past and present that they enjoy.
** The progressive mind want so shape the future to make that future a better place for us all.
* The Collingridge dilemma describes the challenge of regulating new technologies.
** Early stage: When a technology is new, its future impact is unclear, making regulation difficult.
** Later stage: Once the technology is widely adopted, it becomes harder to control or modify.

===Why Bother?===
Some jobs are too risky.
Some tasks are too mind-numbing.
Some conditions are too toxic.
Some decisions are too complex.

AI should be our answer to protect, preserve, and uplift human potential, not disregard or displace it.

===What is Intelligence===
* Here Intelligence relates primarily to [[Identity]] [[Knowledge]], but that has many general aspects.
* In 1950 Turing asked "Can Machines Think".<ref>A. M. Turing, ''COMPUTING MACHINERY AND INTELLIGENCE'' 1950-10-01 https://academic.oup.com/mind/article/LIX/236/433/986238</ref>
* People have been working on artificial intelligence starting with a conference at Dartmouth College in the 1950s. There, computer scientists and mathematicians set a very ambitious goal: to develop computers that could understand language, translate it, see patterns, and more.
* Marvin Minsky was teaching undergrad courses in 1964 using the book "Computers and Thought."<ref>Edward Feigenbaum, & Julian Feldman, joint ed ''Computers and Thought'' https://archive.org/details/computersthought00feig</ref>
* in 2023 Bill Gates in "AI is about to completely change how you use computers" said<ref>Bill Gates ''AI is about to completely change how you use computers'' https://www.gatesnotes.com/AI-agents</ref> "I still love software as much today as I did when Paul Allen and I started Microsoft. But—even though it has improved a lot in the decades since then—in many ways, software is still pretty dumb.:
* Some people think that [[Artificial Intelligence]] must needs be cold or sterile, but we have plenty of [[Evidence]] that is not so.
** Elisa was the first but [https://arstechnica.com/information-technology/2023/12/real-humans-appeared-human-63-of-the-time-in-recent-turing-test-ai-study/ it best chat gpt so I guess the turning test isn't so useful]
** Alexa works on Amazon echo devices and is constantly updated so there is nothing stable to compare.
** GPT was getting some press with GPT-2 included in Technology Review tagged it as a break-through technology in 2021<ref>Will Douglas, '' GPT-3'' Technology Review '''Vol 124''' No 2 (2021-03) pp. 34-35</ref>, but nothing managed to get public [[Attention]] until... {See below for more on Open AI}
** Chat-GPT which was the first publicly accessible product that was not completely cringe-worthy on 2022-11-30.
* Is [[Artificial Intelligence]] even [https://www.linkedin.com/pulse/machine-intelligence-artificial-sean-manion-phd-oz91e/ artificial]? Or is it just yet another way to create a balance between [[Chaos and Order]].
* Determining what [[Artificial Intelligence]] actually is and whether is is like human intelligence turns out to be a impossible task. Ray Kurzweil said<ref>Ray Kurzweil, ''The Singularity is Nearer''</ref><blockquote>The reason a well-defined empirical test is necessary is that ... humans have a powerful tendency to redefine whatever artificial intelligence achieves is not ''really'' so hard in hindsight. This is often referred to as the "AI effect". ... Intuitively [the hallucinations and odd answers] feels like a problem. It's tempting to think that Watson "ought" to reason like humans do. But i would propose that this is superstition. In the real world, what matters is how an intelligent being acts.</blockquote>
===Origin===
Over the summer of 1956 a small but illustrious group gathered at Dartmouth College in New Hampshire; it included Claude Shannon, the begetter of information theory, and Herbert A. Simon, the only person ever to win both the Nobel Memorial Prize in Economic Sciences awarded by the Royal Swedish Academy of Sciences and the Turing Award awarded by the Association for Computing Machinery. They had been called together by a young researcher, John McCarthy, who wanted to discuss “how to make machines use language, form abstractions and concepts” and “solve kinds of problems now reserved for humans”. It was the first academic gathering devoted to what McCarthy dubbed “artificial intelligence”. And it set a template for the field’s next 60-odd years in coming up with no advances on a par with its ambitions.<ref>John McCarthy, M. Minsky, N. Rochester, C.E Shannon, ''A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence'' (1955-08) http://raysolomonoff.com/dartmouth/boxa/dart564props.pdf</ref>

During 1955 and 1956 computer scientist and cognitive psychologist Allen Newell, political scientist, economist and sociologist Herbert A. Simon, and systems programmer John Clifford Shaw, all working at the Rand Corporation in Santa Monica, California, developed the Logic Theorist, the first program deliberately engineered to mimic the problem solving skills of a human being. They decided to write a program that could prove theorems in the propositional calculus like those in Principia Mathematica by Alfred North Whitehead and Bertrand Russell.<ref>''Newell, Simon & Shaw Develop the First Artificial Intelligence Program'' https://www.historyofinformation.com/detail.php?id=742</ref>

In 1956 John von Neumann agreed to present a discussion on "The Computer and the Brain" as the Yale Silliman lecture. He was too ill to complete the full gamut of lectures but did get enough presented to be published by Yale.<ref>John von Neumann, ''The Computer and the Brain'' (1958) Yale UP ISBN 0300024150</ref> Perhaps the most interesting part is the discussion of the difference between arithmetic and logic. He noted that only a few bits of information were available to make a choice between options, but that the computer's arithmetic was growing to every more bits, now 64 in most machines, to give sufficient precision to make a choice. Only machines defined after 2020 used ultra-low precision (2–4 bits) primarily for inference, not training. IBM researchers presented techniques for 4-bit quantization at NeurIPS 2020, focusing on reducing memory and compute costs while maintaining accuracy for deep neural networks

The first international conference was "Mechanisation of Thought Processes"<ref>National Physical Laboratory at Teddington,''Mechanisation of Thought Processes'' https://aitopics.org/download/aiclassics:98875BCD</ref> was held in 1958 with presentations by Minsky and John McCarthy.

===Perceptron===
Also called a McCulloch-Pitts [[Neuron]] is an attempt to mimic an animal [[Neuron]], originally with a purpose built computer system. While it was first described in 1947, it was not realized until Frank Rosenblatt built is Mark I computer using perceptrons as a primary physical element. He continued to build larger machines with more layer until his early death in 1971. Minskey and Papert wrote an entire book<ref> M. Minsky & S. Papert, ''Perceptrons'' (1987-12-28) ISBN 978-0262631112</ref> where they purported to show that it was a dead-end idea which basically blocked progress on them til the 1980's. The Large Language Models that became popular in the early 2020's are based on Multi-Level Perceptrons (MLP).

===End Game===
Now i am become death, the destroyer of worlds. Quoted by Robert Oppenheimer at the completion of the Atomic bomb and in the movie "Ex Machina" from the Bhagavad Gita.

Perhaps a good question to ask is, "Can Intelligence Be Separated from the Body?"<ref>Oliver Whang, ''Can Intelligence Be Separated From the Body?'' New York Times (2023-04-11) https://www.nytimes.com/2023/04/11/science/artificial-intelligence-body-robots.html?</ref><blockquote>Maybe the mind is like a video game controller, moving the body around the world, taking it on joy rides. Or maybe the body manipulates the mind with hunger, sleepiness and anxiety, something like a river steering a canoe. Is the mind like electromagnetic waves, flickering in and out of our light-bulb bodies? Or is the mind a car on the road? A ghost in the machine?</blockquote>

Perhaps a better question to ask is, "Can Intelligence Be Separated from a persistent Identifier"?<blockquote>The [[Identifier]]s used with AI in 2023 are basically just service [[Identifier]]s. in 2023 at the beginning of the general availability of the deployments of AIs that did not embarrass their creators, was also the time that context was provided to the AI to help it understand a thread of a conversation. Before that every query was de novo, without context. But still there is not a persistent history of conversations, or any concept that that is an "I" there to establish a history, which is still the basis of trust. Perhaps we must wait until we are able to crate an AI that persists an [[Identity]], we, as individuals, will be able to establish trust with it, as an individual. </blockquote>

==Problems==
* Humans seem to not be able to use [[Intelligent Design]] alone to fashion [[Artificial Intelligence]]. In fact researchers continue to go back to nature and human functioning for inspiration and even for algorithms. In other words, our understanding of the meaning of the term [[Artificial Intelligence]] has been and will continue to be evolving over time.
* Nature itself since the Big Bang seems to be based on [[Self-organization]] for its framework.
* Training an [[Artificial Intelligence]] with human behaviors will result in unacceptable behavior by the [[Artificial Intelligence]]. The most common complaint is real, or imagined, or invented, bias against one group or another.
** Microsoft released Tay,<ref>Peter Lee, Learning from Tay’s introduction. (2016-05-25) https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/</ref> a web robot (bot) to respond to tweets and chats<ref>Sarah Parez, ''Microsoft silences its new A.I. bot Tay, after Twitter users teach it racism.'' (2018-03-24) Tech Crunch https://techcrunch.com/2016/03/24/microsoft-silences-its-new-a-i-bot-tay-after-twitter-users-teach-it-racism/</ref> in May 2016. The result was disastrous as Tay learned to be racist from its all-too-human trainers and was shut down in days.
** Google has been plagued with reports and legal action on its search results nearly continuously since it was introduced; in 2019 the president, Donald Trump, accusing it of favoritism to leftist causes.<ref>Farhad Manjoo, ''Search Bias, Blind Spots And Google.'' (2018-08-31) New York Times p. B1</ref> Researcher Safiya U. Noble has written a book<ref>Safiya U. Noble, ''Algorithms of Oppression: How Search Engines Reinforce Racism.'' (2018-02-20) ISBN 978-1479849949</ref> mostly complaining that all-too-human programmers injected their own prejudices into their work. What else could he expect of humans, to rise above them selves, whatever that might mean in terms of free speech or freedom of religion?
** The page [[Right to be Forgotten]] describes an effort in Europe to teach the search engines to with-hold information about people that they don't like by collecting all of the information that they don't like. Be careful what you ask your [[Artificial Intelligence]] to do for you; it might just spill the beans some day, perhaps under court order.
** In the movie "Blade Runner 2049" the protagonist's AI girl friend asks to have her memory wiped so that the police cannot compel her to testify against him. One would hope that our future AIs will be that compassionate towards us; but that kind of compassionate behavior will probably be illegal.
===Personhood===
This is a really bad idea. It's solving a problem from the last generation of wallets when we need something that will work when the wallet is just an AI agent that can mimic anything thrown at it. Face it - the cred even today comes from a wallet that needs to prove its own creds before it can say anything about who is (or is not) at the controls. Face it - the internet of the future is just ai agents talking to each other.
* The development of full [[Artificial Intelligence]] could spell the end of the human race. - Stephan Hawking BBC Interview 2014. There are lots of dystopian futures depicted by science fiction authors. Some, like the Terminator series seem unlikely. The scenario in the movie Gattaca where humans were separated into categories based on the genome has already been enabled in the Personhood credential<ref>Pymnts, ''AI’s Human Mimicry Spurs ‘Personhood Credential’ Proposal'' 2024-09-24 https://www.pymnts.com/digital-identity/2024/ais-human-mimicry-spurs-personhood-credential-proposal/</ref> which would allow machines to determine by themselves who is qualified to be classified as human.
* Descartes claimed "I think therefore I am". But it seems that we need to be careful about applying that idea to [[Artificial Intelligence]]. Although his philosophy is no longer considered to be valid even for humans.
* [https://workshop.vennfactory.com/p/personhood-the-killer-credential Personhood: The Killer Credential?] Digging deeper into a very human challenge - Eve Maler 2024-09-16

===Vulnerabilities===
* 2024-08-20 [https://www.w3.org/reports/ai-web-impact/#safety-and-security AI & the Web: Understanding and managing the impact of Machine Learning models on the Web] from the W3C
* [https://home.treasury.gov/news/press-releases/jy2212U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector] U.S. Department of the Treasury 2024-03-27
* [https://arxiv.org/pdf/2402.00888.pdf Security and Privacy Challenges of Large Language Models: A Survey] (2024-02)
* Input to the model used by the AI come from two sources, each of which can be compromised.
# Training data from the Internet can be tampered with to give bad results.
## For starters, the internet data is typically censored to remove input that would cause embarrassment and [[Conduct Risk]] for the sponsors.
## Some locales will censor data to make themselves look good or promote some social point of view of the government of of the industry.
# The input from a single user can cause unexpected changes to later results.
## The early AI release were compromised with a variety of hate speech that quickly resulted in really nasty output from the AI.
## Artists seeking to protect their intellectual property from later output have learned to put small changes in the images that cause them to be mislabeled or worse.<ref>Melissa Hakkiliia, ''This new data-poisoning tool lets artists fight back against generative AI'' MIT Technology Review 127 No 1 (2024-01) p 7=8</ref>
* Giving a AI a goal may result in unexpected and undesired solutions.
** Bruce Schneier described<ref>Bruce Schneier, ''Hackers Used to Be Humans. Soon, AIs Will Hack Humanity'' Wired (2021-04-18) https://www.wired.com/story/opinion-hackers-used-to-be-humans-soon-ais-will-hack-humanity/?bxid=5c5b250d24c17c67f8640083</ref> AI to be Like crafty genies. They will grant our wishes and then hack them, exploiting our social, political, and economic systems like never before.

See the wiki on [[AI Threat Model]]s for more details on this subject.

===Uniquely Human===
Neither the Turning Test, nor any other known test, is designed to determine if an [[Artificial Intelligence]] is really human. The best way to understand what natural intelligence is would be to look at the reason human evolved to have such a large brain. There appear to be these principle requirements:<ref>Edward O. Wilson, ''The Meaning of Human Existences'' Liveright (2014) Chapter 2 ISBN 9780871401007</ref>
# Necessity for a large brain is the changing [[Ecosystem]] that a small remnant of Homo Sapiens living in the south of Africa needed in order to survive
# Changing locations to track new sources of protein and the switch from plant to animal protein.
# Empathy to be able to understand what is in someone else's mind to deduce if they notice you and if they think you are targeted by them as prey.
The results of these were two processes that served to enable survival in such varying [[Ecosystem]]s/
# Passing learning from one generation to the next.
# An extended adolescence with mothers that ceased being fecund to tend to their existing children and the extended life of grandparents to help with stories and other techniques to bring the knowledge of the tribe to the growing child.
The search for causes that were more complex than the ability of genes to carry information, such as the migratory pattern of birds is described with more detail in the wiki page on [[Chance and Necessity]].

* [https://www.linkedin.com/pulse/machine-intelligence-artificial-part-4-cybernetics-sean-manion-phd-kfi0e Machine Intelligence is not Artificial - Part 4: Cybernetics and Norbert Wiener]<blockquote>What is of particular note is not simply the existence of cybernetics as a discipline dating back to the 1940s, but the broad popularity Wiener's book and the topic received not only from scientists and engineers, but also the general public. His 1948 book went through four printings in the first 6 months and was a best seller, rivaling the contemporary Kinsey Report in popularity over the following decade. This was despite this book, ostensibly for a lay audience, being extremely dense with many chapters full of mathematical equations. One of the chapters in the original version, "The Computing Machines and the Nervous System," makes it clear that Wiener's mention of humans AND machines in the book's subtitle was pointing towards the comparison of these systems, making this book one of the early popularizers of the idea of "thinking machines" (but by no means the only one). This was two years before Alan Turing's 1950 paper "Computing Machinery and Intelligence," which gave us the Turing test for AI.</blockquote>

* Human-centered AI - [https://aiindex.stanford.edu/report/?utm_source=Stanford+HAI&utm_campaign=525788e5b7-AI_INDEX_2024_EMAIL_CAMPAIGN&utm_medium=email&utm_term=0_-525788e5b7-%5BLIST_EMAIL_ID%5D&mc_cid=525788e5b7&mc_eid=e55581d782 THE AI INDEX REPORT - Measuring trends in AI] 2024 HAI Stanford

===Explaining AI===
Most users want to know '''WHY''' a decision was make or an action was performed. The article "Solving for Why"<ref> Marina Kraskovsky, ''Solving for Why'' '''CACM 65''' No 2 (2202-02) pp 11ff</ref> is a focus on [[Causality]] to help researchers overcome shortcomings that have bedeviled more traditional approaches to AI. It is not enough to predict things, a good AI will need to solves things and explain why the choice was make if the decision is to have any support by the human population. People will ask if the decision was fair and will second guess the choices make by the AI unless they understand and accept the decision. Today AI decision are fragile, a slight change in the environment may have drastic changes in outcomes. This may well be inexplicable to humans. It will be insufficient to convince experts in the field about the choices made. It is apparent in 2022 that expert opinions about COVID have not been very popular. Even more evidence will be required of a AI decision. This will be particularly true when the decision is costly in any way. Humans like to have choices, which may require the AI to offer choices. That will have the benefit of giving the humans some feeling that they are in control. But beware of "Hobson's choice". That was a stable master at Oxford that was required to give the students a choice of horses, but selected the options so that there was no real choice.

===Creativity===
the idea that early large language models (LLMs) showed more creative writing, and that this quality degraded as rules, alignment, and post‑training constraints were added. I’ll separate what is documented, what is strongly suggested by reporting, and what remains a hypothesis, and I’ll anchor claims to explicit sources; especially the recent work of Jasmine Sun reports a striking consensus among AI researchers and writers she interviewed:
* Earlier models (especially GPT‑2–era systems) often produced unexpected, playful, and strange continuations.
* Modern models are more fluent and correct—but less surprising and less interesting to read.

Sun explicitly reports examples from researchers and poets who experimented with GPT‑2, describing outputs that wandered, took risks, and made odd but evocative associations that newer models no longer produce. This is not nostalgia alone: Sun notes that AI executives themselves admit they have not released a model that writes well, despite enormous gains in capability.

Jasmine Sun is currently The Atlantic’s clearest voice arguing that AI degrades creativity by eliminating stakes, judgment, and responsibility, even as it perfects surface‑level fluency.<ref>Sasmine Sun, ''The Human Skill That Eludes AI'' The Atlantic 2026-03-17</ref><ref>Hard Fork ''AI-Washing Layoff? + Why LLMs Can't Write Well + Tokenmaxxing 2026-03-20</ref>

===Paradox===
Moravec’s Paradox describes our experience that it is hard to teach machines to do things that are easy for most humans (walking, for example) but comparatively easy to teach them things that are challenging for most humans (chess comes to mind).

Polanyi’s Paradox is named for the philosopher Michael Polanyi (1891-1976), who developed the concept of “tacit knowledge”: Central to Michael Polanyi’s thinking was the belief that creative acts (especially acts of discovery) are shot-through or charged with strong personal feelings and commitments (hence the title of his most famous work Personal Knowledge). Arguing against the then dominant position that science was somehow value-free, Michael Polanyi sought to bring into creative tension a concern with reasoned and critical interrogation with other, more ‘tacit’, forms of knowing. Polanyi’s argument was that the informed guesses, hunches and imaginings that are part of exploratory acts are motivated by what he describes as ‘passions’. They might well be aimed at discovering ‘truth’, but they are not necessarily [[Formal Model]]s that can be stated in propositional or formal terms. As Michael Polanyi (1967: 4) wrote in The Tacit Dimension, we should start from the fact that ‘we can know more than we can tell‘.<ref>MARK K. SMITH, “MICHAEL POLANYI AND TACIT KNOWLEDGE” AT INFED</ref>

* For synonyms to tacit knowledge see the wiki page on [[Common Sense]].
* It seems, in 2023, that generative AI is able to overcome the lack of imagination with simple random guessing where knowledge is lacking.

===Fairness===
* What is called "Algorithmic Bias" today is just the result of training any [[Artificial Intelligence]] using any data based on human behavior which typically is biased.
* We now expect more ''Fairness'' from our [[Artificial Intelligence]] as we expect from normal humans.
* "The burgeoning field of algorithmic fairness, part of the much broader field of responsible computing, is aiming to remedy the situation"<ref> Marina Krakovsky, '' Formalizing Fairness'' '''CACM 65''' No. 8 (2022-08) p 11ff.</ref> which is the result of using available data.
* Scientists have found that the English word "fairness" has many meanings, and do they would prefer terms like technical disparities, that they can define precisely. The problem with that is that what society demands is "fairness". Which means that scientists could never deliver what society demands.
* As was demonstrated in the COMPASS program to predict recidivism in convicts, it is not possible create a single solution that satisfies more than one metric for "fairness".
* Explainable Fairness means can you explain it to a judge so that they can make a determination. -- Cathy O'Neil ORCAA
* Every provider wants you to think that they are meeting the presidents EO, like the guys pitching Zero Trust Network Access (ZTNA) - do not be confused, this meets no level of [[Least Privilege]] or any other attempt to limit access to people that have earned access. It is nothing but the network folk trying to climb on band wagon.
* [https://www.readynez.com/en/blog/understanding-fair-ai-a-simple-guide/ Understanding Fair AI: A Simple Guide!]
* See the wiki on [[Fairness Accountability Transparency Ethics]] for more details on this topic.

===Disruption===
We can now say what GPT can replace - shallow human conversation and mechanical activity like law or medicine clerks perform. A GPT in a cocktail party could be wittier than anyone else. A GPT could do better than a recent grad of most law or medical schools at citing existing law or medical practice. Is anyone surprised? AI is moving up the food chain from one mechanical task to the next. We learned from aircraft pilot errors that we had to have a checklist to save pilots from error. I guess we could call the current AI a check list of stuff that is already known.

The question about what is left for humans might again be of interest. It was during the industrial revolution when machines replaced human labor. The best that the pundits of the time could suggest it that humans could have more leisure time. Now machines are moving to more tasks. But somehow employment is full in 2022. There is likely to be a reconning some day when machines and their owners will disrupt the system and cause unemployment. The rich and the poor will, yet again, come into conflict and a revolution of some sort will be required.

===Bias===
* Clearly any [[Artificial Intelligence]] is biased by its input data. Social pressures have forced any [[Entity]] that offers an AI to make if conform to the laws and regulations of the state. All countries have proscribed some set of topics or responses as unacceptable. In other words all AI are censored by the training data or the laws of the land. Even the language used for training can create biases.<ref>Antony Chayka and Andrei Sukhov, ''Comparing Chatbots Trained in Different Languages'' '''CACM 66''' no 12 pp, 6-7 (2023-12)</ref><blockquote>In our opinion, the government's position is clearly taken into account in the responses of AI systems in the national language, especially when the creation of AI was funded in the tested country.</blockquote>
* We face the same problems in raising our children. If they are raised in an environment that exhibits bias, it will be hard to change their behavior later in life.<ref>Will Douglas Heaven, ''Six Big Questions for Generative AI: Hope, Hype, Nope'' Technology Review (2024-01) p. 30ff. </ref> But if we use synthetic data to train the AI there will be huge objections as it is not possible for any training material (for either humans or AI) to be objectionable to some subset of humanity.

==Solutions==
The wiki page on [[Trust]] describes how people can get into the habit of trusting business process or technology process that really do not deserve to be trusted. The following is a listing of some of the technological solutions. The question about whether an [[Artificial Intelligence]] is trustworthy is yet to be addressed.

===Levels===
It is to be expected that level of intelligence will be applicable, which the higher levels learning from, or dependent on, simpler levels that start from basic automation and then to the following.<ref>Peter Denning and Ted Lewis, ''An AI Learning Hierarchy'' '''CACM 67''' No 12 (2024-12) p. 24ff.</ref>

# Rule-based system, these were applied to board games early in the computing age, but were not able to Handle more complex games like chess or go where experts were unable to articulate the rules by which they played.
# Supervised Learning
# Unsupervised Learning
# Generative AI
# Reinforcement Learning
# Human-machine interaction
# Aspirational AI

===Governance===
For most of 2023 the focus was on government standards for AI. The idea that trust can be initiated by government oversight is wide-spread, but wrong. Trust comes from a history of fail dealing, that is not something that governments understand or practice. But still the government can, and should, punish crimes, so the legislation should focus on how crimes committed by AI are punished. The death penalty would be one good deterrent. Here is one discussion on what governments can do:
[https://www.linkedin.com/pulse/governance-trust-guy-huntington-cus3c/ AI/bots and National Security].
===Training===
A great deal of effort is directed at the education of young children. In large part education is directed at the production of adults that a useful the the society in which they are expected to function as useful citizens. But their is a sharp disagreement on the subject. According to Ken Robinson, education should have four core purposes: personal, economic, social, and cultural <ref>Ken Robinson, Kate Robinson ''What Is Education For?'' (2022-03-02) https://www.edutopia.org/article/what-education</ref> Perhaps it is time to model the training of AI on the training of humans?

===User Experience===
The real question about the use of the current Large Language Model of [[Artificial Intelligence]] is how the device in our possession displays the interaction with the [[Artificial Intelligence]]. We humans have been reading stories and watching displays of [[Artificial Intelligence]] from the puppet Pinocchio that wanted to be a real boy to Deus ex Machina a movie about an [[Artificial Intelligence]] robot that went rouge. The common thread here is that each was a fictional account that we were able to relate to. Good authors know how to enliven the characters of a story for the reader that drags the reader into the story. The [[User Experience]] of AI for the near term will be interactions through our personal devices, even though the more interesting, and troubling, case will be when the robotic devices are not in our possession, but directly run by some gigantic corporation. The good story tellers will be the ones that grab our [[Attention]], even as is true today.<ref>Hannah H. Kim, ''A.I. is a Fiction'' Wired pp. 9-10 (2023-08) 31-08</ref>

What we must focus on is the UX and its effect on the people that interact with it.

===The Smart Agent===
Now that an A.I. can apply reason to problems, it should be expected that such agents on on the horizon. That prediction was made in 2021, and there seems to be glimmerings of AI appearing in the user agents in our hands in 2023, but it is still all driven by Large Language Models (LLM) in the cloud. Intel is predicting that user devices in 2024 will include and AI. Loading LLM into [[Edge Computing]] devices are needed before AI will work in small devices.

Nvidia has been building GPUs for personal computers for a long time before they were purposed for [[Artificial Intelligence]]. But on 2023-12-14 they were not used for that on personal computers when Intel unveiled [https://www.intel.com/content/www/us/en/newsroom/news/ai-everywhere-core-ultra-5th-gen-xeon-news.html#gs.26279f "Core Ultra and 5th Gen Xeon – that will change how customers and consumers enable AI solutions in the data center, cloud, network, at the edge, and on the PC"] to bring AI to any home computer user with a budget for a new computer.

===Neural Nets===

Still the most popular, they appear to have reached a limit in what they can explain about human intelligence. Perhaps we need to add compassion and [[Common Sense]] back in.

The theory of neural nets has been documented in Quanta <ref>Kevin Hartnett, ''Foundations Built for a General Theory of Neural Networks.'' (2019-01-31) Quanta https://www.quantamagazine.org/foundations-built-for-a-general-theory-of-neural-networks-20190131/</ref> A description of the Turing award to the winners for applying neural nets to solving problem (as opposed to explaining intelligence) was summarized by Neil Savage<ref>Neil Savage, ''Neural Net Worth.'' '''CACM Vol 62''' no 6 p. 12 (2019-06)</ref> who specifically talks about the need for more involvement of humanities in long term solutions.

===Deep Neural Network Accelerators===
Now we come to the logical and nearly the final phase of [[Artificial Intelligence]], the design of one AI by another.<ref>Adam Zewe, ''Accelerating AI tasks while preserving data security'' MIT News (2023-10-30) https://news.mit.edu/2023/accelerating-ai-tasks-while-preserving-data-security-1030</ref> This is probably just the first step to the complete control of the AI [[Ecosystem]] by the AI themselves when the designs become to complex for human understanding.<blockquote>With the proliferation of computationally intensive machine-learning applications, such as chatbots that perform real-time language translation, device manufacturers often incorporate specialized hardware components to rapidly move and process the massive amounts of data these systems demand. Choosing the best design for these components, known as deep neural network accelerators, is challenging because they can have an enormous range of design options. This difficult problem becomes even thornier when a designer seeks to add cryptographic operations to keep data safe from attackers. Now, MIT researchers have developed a search engine that can efficiently identify optimal designs for deep neural network accelerators, that preserve data security while boosting performance.</blockquote>

===Libratus===

Libratus did not use expert domain knowledge, or human data that are specific to [the game domain]. Rather, the AI was able to analyze the game's rules and devise its own strategy. The technology thus could be applied to any number of imperfect-information games. Such hidden information is ubiquitous in real-world strategic interactions, including business negotiation, cybersecurity, finance, strategic pricing and military applications.<ref>Byron Spice, ''CSD's Sandholm, Brown To Receive Minsky Medal.'' (2018-11-06) Carnegie-Mellon News https://www.cmu.edu/news/stories/archives/2018/november/minsky-medal.html</ref>

===Alpha Zero===

After Alpha Go, the first version to use [[Matrix Calculation|Tensor Processing Units]] (TPU) defeating the best human GO play in 2016,<ref>McMillan, Robert. "Google Isn't Playing Games With New Chip". (2016-05-18) The Wall Street Journal. Archived from the original on 29 June 2016. </ref> the Google Deep Mind project Alpha Go was generalized into a far simpler version Alpha Zero<ref>Steven Strogatz, ''One Giant Step for a Chess-Playing Machine.'' (2018-12-26) New York Times https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html</ref> that was just given the rules of the game, whether Go or Chess, and allowed to play against another instance of itself until it had mastered the subject to be able to beat any existing computer model. Garry Kasparov, the world chess said that Alpha Go developed a style of play that "reflects the truth" of the game rather than "the priorities and prejudices of programmers." That pretty much ends the discussion about the superiority of "Intelligent Design" which turns out to be just another myth that humans tell about themselves and their anthropomorphic gods. It also ended the testing of [[Artificial Intelligence]] against humans in board games. There was very little generally newsworthy announcements about [[Artificial Intelligence]] for the next 6 years.

===Open AI===
Open AI Is a research institute founded by Elon Musk and Sam Altman to make AI results available to everyone. Their finding (which echoed that of Deep Mind at Google) in early 2019 was that this is a very bad idea. Like any technology it can ready be turned to evil purposes as reported in Wired.<ref>Alyssa Foote, ''The AI Text Generator That's Too Dangerous to Make Public.'' (219-02-19) Wired Magazine https://www.wired.com/story/ai-text-generator-too-dangerous-to-make-public/</ref> This finding should certain give pause to any company that seeks to use AI for their own purposes. It is too easy for AI to be turned to the purposed of evil intent. Sam Altman wrote an essay ''Moore's Law for Everything''<ref>Sam Altman ''Moore's Law for Everything'' personal blog (2021) https://moores.samaltman.com/</ref> that was reviewed by Ezra Klein.<ref>Ezra Klein ''Sam Altman on the A.I. Revolution, Trillionaires and the Future of Political Power'' NY Time (2021-06-11) https://www.nytimes.com/2021/06/11/opinion/ezra-klein-podcast-sam-altman.html?action=click&module=Opinion&pgtype=Homepage</ref>

Not everyone is captivated by Open AI. Quote from the New Yorker: [https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web "OpenAI’s chatbot offers paraphrases, whereas Google offers quotes. Which do we prefer?"] To be fair, the version called Copilot by Microsoft does provide links to the original documents.

==Ethics==
Most of the effort in protecting the human population from the dangers of [[Artificial Intelligence]] focus on arguments about ethics. [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Quandaries in AI-ML is one of a growing list on YouTube] joined by an ACM webinar on [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Challenges in AI]. What remains unclear is how AI ethics can be separated from [[Censorship]]. Most of the effort at training AI involves a list of taboo subjects. The problem here is that not everyone agrees on what ethics should apply. For example, is abortion in one country to be considered to be murder and in another country just an exercise in free will. It is unclear whether a discussion of abortion methods, suicide or similar health issues is an ethical or a moral discussion. Similar issues are easy to list, such as free speech or social credit. It comes down to a question of whether ethics in AI will ever be any more that just censorship that is inconsistently applied?

===In Healthcare===
* [https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Forange.hosting.lsoft.com%2Ftrk%2Fclick%3Fref%3Dznwrbbrs9_6-2be2bx32c520x081583%26&data=04%7C01%7C%7C1b0cfc8a7db9484bb50308d946e0a1e9%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637618752869025374%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PaRyaYQadjyHbH5wP%2FruYlnD0B5rftISxc8a9dfAJeQ%3D&reserved=0 WHO Releases AI Guidelines for Health Government Technology] 2021-07-12 <blockquote>A new report from the World Health Organization (WHO) offers guidance for the ethical use of artificial intelligence (AI) in the health sector. The six primary principles for the use of AI as set forth in the report are to protect autonomy; promote human well-being, safety, and the public interest; ensure transparency, explainability, and intelligibility; foster [[Responsibility]] and accountability; ensure inclusiveness and equity; and promote responsive and sustainable AI. These principles are intended as a foundation for AI stakeholders, including governments, developers, and society. The report, Ethics and Governance of Artificial Intelligence for Health, finds public- and private-sector collaboration necessary to ensure AI’s accountability.<blockquote>

===Back to the Future===
Brian Christian described "The Alignment Problem"<ref>Brian Christian ''The Alignment Problem'' W. W. Norton (2020-10-06) ISBN 978-0-393-86833-3</ref> as the difficulty that AI has with aligning the results provided by the AI with society's goals. It is worse than you might imagine. The results from AI as of 2020 have not proven to be appropriate. This is, in part, because they are trained on data from social systems that are also deemed to be inappropriate. What most researchers are doing from this point is trying to determine how natual intelligence develops and feeding that back into the AI systems.

The first person to try to understand "The Construction of [[Reality]] in the Child' was Jean Piaget<ref>Jean Piaget ''The Construction of Reality in the Child'' (original 1954) Ballentine ISBN 345-02338-2-165</ref>. The first line of the book is "To understand how the budding intelligence constructs the external world, we must ask whether the child, in its first months of life, conceives and perceives things as we do, as object that have substance, that are permanent and of constant dimensions." Now that NMR brain scanners can image a thought within an infant's mind<ref>Rachel Fritts, ''How do Babies perceive the World?'' MIT News July/August 2021 p 24 ff</ref>, we can actually imagine that the brain really is just the Chinese Room that so perplexed Searle.<ref>James Somers, ''Head Space'' The New Yorker Dec. 6, 2021, p 30 ff.</ref> Searle's objection to [[Artificial Intelligence]] was that it was just a symbol manipulation scheme and that cannot be all that is involved in understanding. Interestingly at the same time Noam Chomsky created a generative linguistic model of language that was nothing more than the manipulation of symbols.<ref>George Lakof, ''The Functionalist's Dilemma'', American Scientist https://www.americanscientist.org/article/the-functionalists-dilemma</ref> As you might imagine, they disagreed often in print.<ref>''Chomsky’s Revolution: An Exchange'' (2002-07-18) https://www.nybooks.com/articles/2002/07/18/chomskys-revolution-an-exchange/</ref> See the wiki page on [[Knowledge]] for more on this dispute. Andy Clark brings together ideas and techniques from robotics, neuroscience, infant psychology, and artificial intelligence. He addresses a broad range of adaptive behaviors, from cockroach locomotion to the role of linguistic artifacts in higher-level thought.<ref>Andy Clark, ''Being There: Putting Brain, Body, and World Together Again'' 1996 https://direct.mit.edu/books/book/3917/Being-TherePutting-Brain-Body-and-World-Together</ref>

And finally, the AI research community is recognizing the turn of AI towards Natural Science and is not too happy about that?<ref>Subbarao Kambhampati, ''AI as (an Ersatz) Natural Science.'' '''CACM 65''' No 9 (2022-09) pp 8-9</ref> Kambhampati reports that "Indeed, in a 2003 lecture, Turing laureate Lesle Lamport sounded alarms about the very possibility of the future of computing belong to biology rather than logic, saying it will lead us to living in a world of homeopathy and faith healing<ref>Leslie Lamport, ''The Future of Computing: Logic or Biology'' (2003-07-23) https://bit.ly/3ahrsaI</ref> Perhaps we should go back to one of the authors of the ultimate book on logic, the Principia Mathematica Alfred North Whitehead<ref>Alfred North Whitehead, ''Principia Mathematica'' Cambridge UP (Second edition 1927)</ref> who said "Metaphysical categories are not dogmatic statements of the obvious, they are tentative formulation of the ultimate generalities. If we consider any scheme of philosophic categories as one complex assertion, and apply it to the logician's alternative, true or false, the answer must be that the scheme is false. The same answer must be given to like question respecting the existing formulate principles of any science."<ref>Alfred North Whitehead, ''Process and Reality'' Free Press (1929)</ref>

==Truth from the AI==
Lying requires belief + intention.
* Deception requires a goal to induce false belief.
* AI has neither. So by definition, it can’t be “lying” or “deceiving” - at least not in the human sense.

Yet the desire to interpret internal model states as belief-like is so strong, researchers are now building probes to "extract" them.
Words like “belief,” “reasoning,” “understanding,” "thought", and “intelligence” are being borrowed from human cognition - without being earned, and devaluing them all in the process. Not just by journalists, or AI researchers, but especially the ones hyping the tech.

Why?
* Not just to replace workers and cut wages - they know if we have no money, we can’t consume. The system collapses.
* This is about reshaping access to knowledge itself. Controlling what you see. How you learn. Who you become. Who you obey. Who you trust.

===Effective Altruism===
Compare this with wiki page on [[Expected Value]].

Under the concept that those with wealth should be careful that the resources that the contribute back to society have a positive impact. In other words, they want to control the way their money is used. An effort to impact AI in this way is underway. <ref>Timnit Gerbu ''Effective Altruism Is Pushing a Dangerous Brand of "AI Safety"'' Wired (2022-11-30) https://www.wired.com/story/effective-altruism-artificial-intelligence-sam-bankman-fried/</ref><blockquote>EA is defined by the Center for Effective Altruism as “an intellectual project, using evidence and reason to figure out how to benefit others as much as possible.” And “evidence and reason” have led many EAs to conclude that the most pressing problem in the world is preventing an apocalypse where an artificially generally intelligent being (AGI) created by humans exterminates us. To prevent this apocalypse, Some of the billionaires who have committed significant funds to this goal include Elon Musk, Vitalik Buterin, Ben Delo, Jaan Tallinn, Peter Thiel, Dustin Muskovitz, and Sam Bankman-Fried, who was one of EA’s largest funders until the recent bankruptcy of his FTX cryptocurrency platform. As a result, all of this money has shaped the field of AI and its priorities in ways that harm people in marginalized groups while purporting to work on “beneficial artificial general intelligence” that will bring techno utopia for humanity. This is yet another example of how our technological future is not a linear march toward progress but one that is determined by those who have the money and influence to control it. </blockquote>

Nietzshe had a very dismal view on the efficacy of altruism<ref>Friedrich Nietzsche. ''The Twilight of the Idols'' no 35</ref> <blockquote>An "altruistic" morality, a morality under which selfishness withers, is in all circumstances a bad sign. This is true of individuals and above all of nations. The best are lacking when selfishness begins to be lacking. Instinctively to select that which is harmful to one, to be lured by "disinterested' motives,—these things almost provide the formula for decadence. "Not to have one's own interests at heart" —this is simply a moral fig-leaf concealing a very different fact, a physiological one, to no longer know how to find what is to my interest."... Disintegration of the instincts!—AII is up with man when he becomes altruistic.— instead of saying ingenuously "I am no longer any good," the lie Of morality in the decadent's mouth says: "Nothing is any good,—life is no good."—A judgment of this kind ultimately becomes a great danger; for it is infectious, and it soon flourishes on the soil of society...</blockquote>

A better answer to AI might be to allow those who work directly with underserved people. The best way to measure the success of a society is to see how it treats the least of its members.<ref>Karen Hao ''A new vision of artificial intelligence for the people'' Technology Review (2022-04-22)</ref><blockquote>This story is the fourth and final part of MIT Technology Review’s series on AI colonialism, the idea that artificial intelligence is creating a new colonial world order. The project is a radical departure from the way the AI industry typically operates. Over the last decade, AI researchers have pushed the field to new limits with the dogma “More is more”: Amass more data to produce bigger models (algorithms trained on said data) to produce better results. The approach has led to remarkable breakthroughs—but to costs as well. Companies have relentlessly mined people for their faces, voices, and behaviors to enrich bottom lines. And models built by averaging data from entire populations have sidelined minority and marginalized communities even as they are disproportionately subjected to the technology’s impacts. Over the years, a growing chorus of experts have argued that these impacts are '''repeating the patterns of colonial history.''' Global AI development, they say, is impoverishing communities and countries that don’t have a say in its development—the same communities and countries already impoverished by former colonial empires.</blockquote>

See the wiki [[Social Justice]] for more details include Herbert Simon on Bounded Rationality.

==Regulation==
The big question is it safe? The simple answer is that new technology always comes with new risks. Antitrust law is the best-known method for controlling enterprises too large to be controlled by the government. Perhaps they can be used to control a super intelligent being of another sort. Perhaps that's all an AI ever is, corporations with computers and imposed limitations.
* 2024-10-24 [https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/ Memorandum on Advancing the United States’ Leadership in Artificial Intelligence; Harnessing Artificial Intelligence to Fulfill National Security Objectives; and Fostering the Safety, Security, and Trustworthiness of Artificial Intelligence]
* NIST issued a call for participation on [https://www.federalregister.gov/documents/2023/11/02/2023-24216/artificial-intelligence-safety-institute-consortium Artificial Intelligence Safety Institute Consortium] 2023-11-02 in the federal resister.
* President Bidin issued a [https://www.whitehouse.gov/ostp/ai-bill-of-rights/ Blueprint for an AI Bill of Rights] (2022-10-04) problems are well documented. In America and around the world, systems supposed to help with patient care have proven unsafe, ineffective, or biased. Algorithms used in hiring and credit decisions have been found to reflect and reproduce existing unwanted inequities or embed new harmful bias and discrimination. Unchecked social media data collection has been used to threaten people’s opportunities, undermine their privacy, or pervasively track their activity—often without their knowledge or consent. Five Principles:
# Safe and Effective Systems - You should be protected from unsafe or ineffective systems.
# Algorithmic Discrimination Protections - You should not face discrimination by algorithms and systems should be used and designed in an equitable way.
# Data Privacy - You should be protected from abusive data practices via built-in protections and you should have [[Agency]] over how data about you is used.
# Notice and Explanation- You should know that an automated system is being used and understand how and why it contributes to outcomes that impact you.
# Human Alternatives, Consideration, and Fallback - You should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter.
* A law that went into effect in New York City in January 2023<ref>Richard Vanderford, ''New York AI Bias Law Prompts Uncertainty'' Wall Street Journal (2022-09-21) </ref> requires companies to audit their artificial intelligence (AI) hiring systems to assess whether they incorporate racial and gender biases. The law holds hiring companies liable for any biases and could impose fines for violations, but lacks clear guidelines for the AI audit process. While the city’s Department of Consumer and Worker Protection has not offered a timeline for when it will publish rules to implement the law, some companies already are taking steps to comply. Said Anthony Habayeb of AI governance software company Monitaur Inc., "Instead of waiting for someone to tell me what to do… I built controls around these applications because I know like with any software, things can and do go wrong."

==References==

<references />

===Other Material===
* See the wiki page on [[Intelligent Goals]].
* See the wiki page on [[Machine Learning]].
* See the wiki page on [[Fairness Accountability Transparency Ethics]]
* See the wiki page on [[Genetic Programming]]

[[Category: Glossary]]
[[Category: Model]]
[[Category: Authentication]]
[[Category: Artificial Intelligence]]

Artificial Intelligence

2026-03-23T00:41:03Z

Tom: /* Truth from the AI */

==Full Title or Meme==
Intelligence without human lived experience or human compassion.

"The future is already here — it's just not very evenly distributed" - William Gibson.

The term [[Artificial Intelligence]] was coined, in part, out of spite. In 1955, the Dartmouth professor John McCarthy assembled a proposal for what became the first-ever conference on AI. McCarthy refused to use the prevailing term, "cybernetics": That would have necessitated inviting Norbert Wiener, then the world's leading cyberneticist, whom he found insufferable. Instead, McCarthy minted a new term that took off in the digital era, taking him along as a founder of the field. <ref>''Fun fact: Cybernetics'' Technology Review (2025-08) p 11</ref>

==Context==
===The Broad concepts===
* There are two great forces both in physics and in human history
** The harnessing of energy which started with the use of animal power to human benefit.
** The harnessing of information which started with philosophy and then Gutenberg's press.
* Likewise there are two great revolutions in human society.
** The Industrial revolution focused on control and use of energy which peaked in the 20th century.
** The Information revolution focused on control and use of information which is peaking in the 21st century.
* The social changes to accompany these revolution had to before the revolutions could complete.
** The educational system needed to be extended to all who work to engineer the control of energy.
** Now we need a second shift to educate most of the workers to control the information needed.
* The use of energy and information in harmony with the reality of the Earth's resources is a challenge.
** The profligate use of energy has altered the environment in ways that threaten human existence.
** The profligate use of information may prove to alter the environment of work to be a bigger threat.
* The struggle between conservative and progress forces in politics has been with us for ever.
** The conservative mind wants to hold onto the privilege of the past and present that they enjoy.
** The progressive mind want so shape the future to make that future a better place for us all.
* The Collingridge dilemma describes the challenge of regulating new technologies.
** Early stage: When a technology is new, its future impact is unclear, making regulation difficult.
** Later stage: Once the technology is widely adopted, it becomes harder to control or modify.

===Why Bother?===
Some jobs are too risky.
Some tasks are too mind-numbing.
Some conditions are too toxic.
Some decisions are too complex.

AI should be our answer to protect, preserve, and uplift human potential, not disregard or displace it.

===What is Intelligence===
* Here Intelligence relates primarily to [[Identity]] [[Knowledge]], but that has many general aspects.
* In 1950 Turing asked "Can Machines Think".<ref>A. M. Turing, ''COMPUTING MACHINERY AND INTELLIGENCE'' 1950-10-01 https://academic.oup.com/mind/article/LIX/236/433/986238</ref>
* People have been working on artificial intelligence starting with a conference at Dartmouth College in the 1950s. There, computer scientists and mathematicians set a very ambitious goal: to develop computers that could understand language, translate it, see patterns, and more.
* Marvin Minsky was teaching undergrad courses in 1964 using the book "Computers and Thought."<ref>Edward Feigenbaum, & Julian Feldman, joint ed ''Computers and Thought'' https://archive.org/details/computersthought00feig</ref>
* in 2023 Bill Gates in "AI is about to completely change how you use computers" said<ref>Bill Gates ''AI is about to completely change how you use computers'' https://www.gatesnotes.com/AI-agents</ref> "I still love software as much today as I did when Paul Allen and I started Microsoft. But—even though it has improved a lot in the decades since then—in many ways, software is still pretty dumb.:
* Some people think that [[Artificial Intelligence]] must needs be cold or sterile, but we have plenty of [[Evidence]] that is not so.
** Elisa was the first but [https://arstechnica.com/information-technology/2023/12/real-humans-appeared-human-63-of-the-time-in-recent-turing-test-ai-study/ it best chat gpt so I guess the turning test isn't so useful]
** Alexa works on Amazon echo devices and is constantly updated so there is nothing stable to compare.
** GPT was getting some press with GPT-2 included in Technology Review tagged it as a break-through technology in 2021<ref>Will Douglas, '' GPT-3'' Technology Review '''Vol 124''' No 2 (2021-03) pp. 34-35</ref>, but nothing managed to get public [[Attention]] until... {See below for more on Open AI}
** Chat-GPT which was the first publicly accessible product that was not completely cringe-worthy on 2022-11-30.
* Is [[Artificial Intelligence]] even [https://www.linkedin.com/pulse/machine-intelligence-artificial-sean-manion-phd-oz91e/ artificial]? Or is it just yet another way to create a balance between [[Chaos and Order]].
* Determining what [[Artificial Intelligence]] actually is and whether is is like human intelligence turns out to be a impossible task. Ray Kurzweil said<ref>Ray Kurzweil, ''The Singularity is Nearer''</ref><blockquote>The reason a well-defined empirical test is necessary is that ... humans have a powerful tendency to redefine whatever artificial intelligence achieves is not ''really'' so hard in hindsight. This is often referred to as the "AI effect". ... Intuitively [the hallucinations and odd answers] feels like a problem. It's tempting to think that Watson "ought" to reason like humans do. But i would propose that this is superstition. In the real world, what matters is how an intelligent being acts.</blockquote>
===Origin===
Over the summer of 1956 a small but illustrious group gathered at Dartmouth College in New Hampshire; it included Claude Shannon, the begetter of information theory, and Herbert A. Simon, the only person ever to win both the Nobel Memorial Prize in Economic Sciences awarded by the Royal Swedish Academy of Sciences and the Turing Award awarded by the Association for Computing Machinery. They had been called together by a young researcher, John McCarthy, who wanted to discuss “how to make machines use language, form abstractions and concepts” and “solve kinds of problems now reserved for humans”. It was the first academic gathering devoted to what McCarthy dubbed “artificial intelligence”. And it set a template for the field’s next 60-odd years in coming up with no advances on a par with its ambitions.<ref>John McCarthy, M. Minsky, N. Rochester, C.E Shannon, ''A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence'' (1955-08) http://raysolomonoff.com/dartmouth/boxa/dart564props.pdf</ref>

During 1955 and 1956 computer scientist and cognitive psychologist Allen Newell, political scientist, economist and sociologist Herbert A. Simon, and systems programmer John Clifford Shaw, all working at the Rand Corporation in Santa Monica, California, developed the Logic Theorist, the first program deliberately engineered to mimic the problem solving skills of a human being. They decided to write a program that could prove theorems in the propositional calculus like those in Principia Mathematica by Alfred North Whitehead and Bertrand Russell.<ref>''Newell, Simon & Shaw Develop the First Artificial Intelligence Program'' https://www.historyofinformation.com/detail.php?id=742</ref>

In 1956 John von Neumann agreed to present a discussion on "The Computer and the Brain" as the Yale Silliman lecture. He was too ill to complete the full gamut of lectures but did get enough presented to be published by Yale.<ref>John von Neumann, ''The Computer and the Brain'' (1958) Yale UP ISBN 0300024150</ref> Perhaps the most interesting part is the discussion of the difference between arithmetic and logic. He noted that only a few bits of information were available to make a choice between options, but that the computer's arithmetic was growing to every more bits, now 64 in most machines, to give sufficient precision to make a choice. Only machines defined after 2020 used ultra-low precision (2–4 bits) primarily for inference, not training. IBM researchers presented techniques for 4-bit quantization at NeurIPS 2020, focusing on reducing memory and compute costs while maintaining accuracy for deep neural networks

The first international conference was "Mechanisation of Thought Processes"<ref>National Physical Laboratory at Teddington,''Mechanisation of Thought Processes'' https://aitopics.org/download/aiclassics:98875BCD</ref> was held in 1958 with presentations by Minsky and John McCarthy.

===Perceptron===
Also called a McCulloch-Pitts [[Neuron]] is an attempt to mimic an animal [[Neuron]], originally with a purpose built computer system. While it was first described in 1947, it was not realized until Frank Rosenblatt built is Mark I computer using perceptrons as a primary physical element. He continued to build larger machines with more layer until his early death in 1971. Minskey and Papert wrote an entire book<ref> M. Minsky & S. Papert, ''Perceptrons'' (1987-12-28) ISBN 978-0262631112</ref> where they purported to show that it was a dead-end idea which basically blocked progress on them til the 1980's. The Large Language Models that became popular in the early 2020's are based on Multi-Level Perceptrons (MLP).

===End Game===
Now i am become death, the destroyer of worlds. Quoted by Robert Oppenheimer at the completion of the Atomic bomb and in the movie "Ex Machina" from the Bhagavad Gita.

Perhaps a good question to ask is, "Can Intelligence Be Separated from the Body?"<ref>Oliver Whang, ''Can Intelligence Be Separated From the Body?'' New York Times (2023-04-11) https://www.nytimes.com/2023/04/11/science/artificial-intelligence-body-robots.html?</ref><blockquote>Maybe the mind is like a video game controller, moving the body around the world, taking it on joy rides. Or maybe the body manipulates the mind with hunger, sleepiness and anxiety, something like a river steering a canoe. Is the mind like electromagnetic waves, flickering in and out of our light-bulb bodies? Or is the mind a car on the road? A ghost in the machine?</blockquote>

Perhaps a better question to ask is, "Can Intelligence Be Separated from a persistent Identifier"?<blockquote>The [[Identifier]]s used with AI in 2023 are basically just service [[Identifier]]s. in 2023 at the beginning of the general availability of the deployments of AIs that did not embarrass their creators, was also the time that context was provided to the AI to help it understand a thread of a conversation. Before that every query was de novo, without context. But still there is not a persistent history of conversations, or any concept that that is an "I" there to establish a history, which is still the basis of trust. Perhaps we must wait until we are able to crate an AI that persists an [[Identity]], we, as individuals, will be able to establish trust with it, as an individual. </blockquote>

==Problems==
* Humans seem to not be able to use [[Intelligent Design]] alone to fashion [[Artificial Intelligence]]. In fact researchers continue to go back to nature and human functioning for inspiration and even for algorithms. In other words, our understanding of the meaning of the term [[Artificial Intelligence]] has been and will continue to be evolving over time.
* Nature itself since the Big Bang seems to be based on [[Self-organization]] for its framework.
* Training an [[Artificial Intelligence]] with human behaviors will result in unacceptable behavior by the [[Artificial Intelligence]]. The most common complaint is real, or imagined, or invented, bias against one group or another.
** Microsoft released Tay,<ref>Peter Lee, Learning from Tay’s introduction. (2016-05-25) https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/</ref> a web robot (bot) to respond to tweets and chats<ref>Sarah Parez, ''Microsoft silences its new A.I. bot Tay, after Twitter users teach it racism.'' (2018-03-24) Tech Crunch https://techcrunch.com/2016/03/24/microsoft-silences-its-new-a-i-bot-tay-after-twitter-users-teach-it-racism/</ref> in May 2016. The result was disastrous as Tay learned to be racist from its all-too-human trainers and was shut down in days.
** Google has been plagued with reports and legal action on its search results nearly continuously since it was introduced; in 2019 the president, Donald Trump, accusing it of favoritism to leftist causes.<ref>Farhad Manjoo, ''Search Bias, Blind Spots And Google.'' (2018-08-31) New York Times p. B1</ref> Researcher Safiya U. Noble has written a book<ref>Safiya U. Noble, ''Algorithms of Oppression: How Search Engines Reinforce Racism.'' (2018-02-20) ISBN 978-1479849949</ref> mostly complaining that all-too-human programmers injected their own prejudices into their work. What else could he expect of humans, to rise above them selves, whatever that might mean in terms of free speech or freedom of religion?
** The page [[Right to be Forgotten]] describes an effort in Europe to teach the search engines to with-hold information about people that they don't like by collecting all of the information that they don't like. Be careful what you ask your [[Artificial Intelligence]] to do for you; it might just spill the beans some day, perhaps under court order.
** In the movie "Blade Runner 2049" the protagonist's AI girl friend asks to have her memory wiped so that the police cannot compel her to testify against him. One would hope that our future AIs will be that compassionate towards us; but that kind of compassionate behavior will probably be illegal.
===Personhood===
This is a really bad idea. It's solving a problem from the last generation of wallets when we need something that will work when the wallet is just an AI agent that can mimic anything thrown at it. Face it - the cred even today comes from a wallet that needs to prove its own creds before it can say anything about who is (or is not) at the controls. Face it - the internet of the future is just ai agents talking to each other.
* The development of full [[Artificial Intelligence]] could spell the end of the human race. - Stephan Hawking BBC Interview 2014. There are lots of dystopian futures depicted by science fiction authors. Some, like the Terminator series seem unlikely. The scenario in the movie Gattaca where humans were separated into categories based on the genome has already been enabled in the Personhood credential<ref>Pymnts, ''AI’s Human Mimicry Spurs ‘Personhood Credential’ Proposal'' 2024-09-24 https://www.pymnts.com/digital-identity/2024/ais-human-mimicry-spurs-personhood-credential-proposal/</ref> which would allow machines to determine by themselves who is qualified to be classified as human.
* Descartes claimed "I think therefore I am". But it seems that we need to be careful about applying that idea to [[Artificial Intelligence]]. Although his philosophy is no longer considered to be valid even for humans.
* [https://workshop.vennfactory.com/p/personhood-the-killer-credential Personhood: The Killer Credential?] Digging deeper into a very human challenge - Eve Maler 2024-09-16

===Vulnerabilities===
* 2024-08-20 [https://www.w3.org/reports/ai-web-impact/#safety-and-security AI & the Web: Understanding and managing the impact of Machine Learning models on the Web] from the W3C
* [https://home.treasury.gov/news/press-releases/jy2212U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector] U.S. Department of the Treasury 2024-03-27
* [https://arxiv.org/pdf/2402.00888.pdf Security and Privacy Challenges of Large Language Models: A Survey] (2024-02)
* Input to the model used by the AI come from two sources, each of which can be compromised.
# Training data from the Internet can be tampered with to give bad results.
## For starters, the internet data is typically censored to remove input that would cause embarrassment and [[Conduct Risk]] for the sponsors.
## Some locales will censor data to make themselves look good or promote some social point of view of the government of of the industry.
# The input from a single user can cause unexpected changes to later results.
## The early AI release were compromised with a variety of hate speech that quickly resulted in really nasty output from the AI.
## Artists seeking to protect their intellectual property from later output have learned to put small changes in the images that cause them to be mislabeled or worse.<ref>Melissa Hakkiliia, ''This new data-poisoning tool lets artists fight back against generative AI'' MIT Technology Review 127 No 1 (2024-01) p 7=8</ref>
* Giving a AI a goal may result in unexpected and undesired solutions.
** Bruce Schneier described<ref>Bruce Schneier, ''Hackers Used to Be Humans. Soon, AIs Will Hack Humanity'' Wired (2021-04-18) https://www.wired.com/story/opinion-hackers-used-to-be-humans-soon-ais-will-hack-humanity/?bxid=5c5b250d24c17c67f8640083</ref> AI to be Like crafty genies. They will grant our wishes and then hack them, exploiting our social, political, and economic systems like never before.

See the wiki on [[AI Threat Model]]s for more details on this subject.

===Uniquely Human===
Neither the Turning Test, nor any other known test, is designed to determine if an [[Artificial Intelligence]] is really human. The best way to understand what natural intelligence is would be to look at the reason human evolved to have such a large brain. There appear to be these principle requirements:<ref>Edward O. Wilson, ''The Meaning of Human Existences'' Liveright (2014) Chapter 2 ISBN 9780871401007</ref>
# Necessity for a large brain is the changing [[Ecosystem]] that a small remnant of Homo Sapiens living in the south of Africa needed in order to survive
# Changing locations to track new sources of protein and the switch from plant to animal protein.
# Empathy to be able to understand what is in someone else's mind to deduce if they notice you and if they think you are targeted by them as prey.
The results of these were two processes that served to enable survival in such varying [[Ecosystem]]s/
# Passing learning from one generation to the next.
# An extended adolescence with mothers that ceased being fecund to tend to their existing children and the extended life of grandparents to help with stories and other techniques to bring the knowledge of the tribe to the growing child.
The search for causes that were more complex than the ability of genes to carry information, such as the migratory pattern of birds is described with more detail in the wiki page on [[Chance and Necessity]].

* [https://www.linkedin.com/pulse/machine-intelligence-artificial-part-4-cybernetics-sean-manion-phd-kfi0e Machine Intelligence is not Artificial - Part 4: Cybernetics and Norbert Wiener]<blockquote>What is of particular note is not simply the existence of cybernetics as a discipline dating back to the 1940s, but the broad popularity Wiener's book and the topic received not only from scientists and engineers, but also the general public. His 1948 book went through four printings in the first 6 months and was a best seller, rivaling the contemporary Kinsey Report in popularity over the following decade. This was despite this book, ostensibly for a lay audience, being extremely dense with many chapters full of mathematical equations. One of the chapters in the original version, "The Computing Machines and the Nervous System," makes it clear that Wiener's mention of humans AND machines in the book's subtitle was pointing towards the comparison of these systems, making this book one of the early popularizers of the idea of "thinking machines" (but by no means the only one). This was two years before Alan Turing's 1950 paper "Computing Machinery and Intelligence," which gave us the Turing test for AI.</blockquote>

* Human-centered AI - [https://aiindex.stanford.edu/report/?utm_source=Stanford+HAI&utm_campaign=525788e5b7-AI_INDEX_2024_EMAIL_CAMPAIGN&utm_medium=email&utm_term=0_-525788e5b7-%5BLIST_EMAIL_ID%5D&mc_cid=525788e5b7&mc_eid=e55581d782 THE AI INDEX REPORT - Measuring trends in AI] 2024 HAI Stanford

===Explaining AI===
Most users want to know '''WHY''' a decision was make or an action was performed. The article "Solving for Why"<ref> Marina Kraskovsky, ''Solving for Why'' '''CACM 65''' No 2 (2202-02) pp 11ff</ref> is a focus on [[Causality]] to help researchers overcome shortcomings that have bedeviled more traditional approaches to AI. It is not enough to predict things, a good AI will need to solves things and explain why the choice was make if the decision is to have any support by the human population. People will ask if the decision was fair and will second guess the choices make by the AI unless they understand and accept the decision. Today AI decision are fragile, a slight change in the environment may have drastic changes in outcomes. This may well be inexplicable to humans. It will be insufficient to convince experts in the field about the choices made. It is apparent in 2022 that expert opinions about COVID have not been very popular. Even more evidence will be required of a AI decision. This will be particularly true when the decision is costly in any way. Humans like to have choices, which may require the AI to offer choices. That will have the benefit of giving the humans some feeling that they are in control. But beware of "Hobson's choice". That was a stable master at Oxford that was required to give the students a choice of horses, but selected the options so that there was no real choice.

===Creativity===
the idea that early large language models (LLMs) showed more creative writing, and that this quality degraded as rules, alignment, and post‑training constraints were added. I’ll separate what is documented, what is strongly suggested by reporting, and what remains a hypothesis, and I’ll anchor claims to explicit sources; especially the recent work of Jasmine Sun reports a striking consensus among AI researchers and writers she interviewed:
* Earlier models (especially GPT‑2–era systems) often produced unexpected, playful, and strange continuations.
* Modern models are more fluent and correct—but less surprising and less interesting to read.

Sun explicitly reports examples from researchers and poets who experimented with GPT‑2, describing outputs that wandered, took risks, and made odd but evocative associations that newer models no longer produce. This is not nostalgia alone: Sun notes that AI executives themselves admit they have not released a model that writes well, despite enormous gains in capability.

Jasmine Sun is currently The Atlantic’s clearest voice arguing that AI degrades creativity by eliminating stakes, judgment, and responsibility, even as it perfects surface‑level fluency.<ref>Sasmine Sun, ''The Human Skill That Eludes AI'' The Atlantic 2026-03-17</ref>

===Paradox===
Moravec’s Paradox describes our experience that it is hard to teach machines to do things that are easy for most humans (walking, for example) but comparatively easy to teach them things that are challenging for most humans (chess comes to mind).

Polanyi’s Paradox is named for the philosopher Michael Polanyi (1891-1976), who developed the concept of “tacit knowledge”: Central to Michael Polanyi’s thinking was the belief that creative acts (especially acts of discovery) are shot-through or charged with strong personal feelings and commitments (hence the title of his most famous work Personal Knowledge). Arguing against the then dominant position that science was somehow value-free, Michael Polanyi sought to bring into creative tension a concern with reasoned and critical interrogation with other, more ‘tacit’, forms of knowing. Polanyi’s argument was that the informed guesses, hunches and imaginings that are part of exploratory acts are motivated by what he describes as ‘passions’. They might well be aimed at discovering ‘truth’, but they are not necessarily [[Formal Model]]s that can be stated in propositional or formal terms. As Michael Polanyi (1967: 4) wrote in The Tacit Dimension, we should start from the fact that ‘we can know more than we can tell‘.<ref>MARK K. SMITH, “MICHAEL POLANYI AND TACIT KNOWLEDGE” AT INFED</ref>

* For synonyms to tacit knowledge see the wiki page on [[Common Sense]].
* It seems, in 2023, that generative AI is able to overcome the lack of imagination with simple random guessing where knowledge is lacking.

===Fairness===
* What is called "Algorithmic Bias" today is just the result of training any [[Artificial Intelligence]] using any data based on human behavior which typically is biased.
* We now expect more ''Fairness'' from our [[Artificial Intelligence]] as we expect from normal humans.
* "The burgeoning field of algorithmic fairness, part of the much broader field of responsible computing, is aiming to remedy the situation"<ref> Marina Krakovsky, '' Formalizing Fairness'' '''CACM 65''' No. 8 (2022-08) p 11ff.</ref> which is the result of using available data.
* Scientists have found that the English word "fairness" has many meanings, and do they would prefer terms like technical disparities, that they can define precisely. The problem with that is that what society demands is "fairness". Which means that scientists could never deliver what society demands.
* As was demonstrated in the COMPASS program to predict recidivism in convicts, it is not possible create a single solution that satisfies more than one metric for "fairness".
* Explainable Fairness means can you explain it to a judge so that they can make a determination. -- Cathy O'Neil ORCAA
* Every provider wants you to think that they are meeting the presidents EO, like the guys pitching Zero Trust Network Access (ZTNA) - do not be confused, this meets no level of [[Least Privilege]] or any other attempt to limit access to people that have earned access. It is nothing but the network folk trying to climb on band wagon.
* [https://www.readynez.com/en/blog/understanding-fair-ai-a-simple-guide/ Understanding Fair AI: A Simple Guide!]
* See the wiki on [[Fairness Accountability Transparency Ethics]] for more details on this topic.

===Disruption===
We can now say what GPT can replace - shallow human conversation and mechanical activity like law or medicine clerks perform. A GPT in a cocktail party could be wittier than anyone else. A GPT could do better than a recent grad of most law or medical schools at citing existing law or medical practice. Is anyone surprised? AI is moving up the food chain from one mechanical task to the next. We learned from aircraft pilot errors that we had to have a checklist to save pilots from error. I guess we could call the current AI a check list of stuff that is already known.

The question about what is left for humans might again be of interest. It was during the industrial revolution when machines replaced human labor. The best that the pundits of the time could suggest it that humans could have more leisure time. Now machines are moving to more tasks. But somehow employment is full in 2022. There is likely to be a reconning some day when machines and their owners will disrupt the system and cause unemployment. The rich and the poor will, yet again, come into conflict and a revolution of some sort will be required.

===Bias===
* Clearly any [[Artificial Intelligence]] is biased by its input data. Social pressures have forced any [[Entity]] that offers an AI to make if conform to the laws and regulations of the state. All countries have proscribed some set of topics or responses as unacceptable. In other words all AI are censored by the training data or the laws of the land. Even the language used for training can create biases.<ref>Antony Chayka and Andrei Sukhov, ''Comparing Chatbots Trained in Different Languages'' '''CACM 66''' no 12 pp, 6-7 (2023-12)</ref><blockquote>In our opinion, the government's position is clearly taken into account in the responses of AI systems in the national language, especially when the creation of AI was funded in the tested country.</blockquote>
* We face the same problems in raising our children. If they are raised in an environment that exhibits bias, it will be hard to change their behavior later in life.<ref>Will Douglas Heaven, ''Six Big Questions for Generative AI: Hope, Hype, Nope'' Technology Review (2024-01) p. 30ff. </ref> But if we use synthetic data to train the AI there will be huge objections as it is not possible for any training material (for either humans or AI) to be objectionable to some subset of humanity.

==Solutions==
The wiki page on [[Trust]] describes how people can get into the habit of trusting business process or technology process that really do not deserve to be trusted. The following is a listing of some of the technological solutions. The question about whether an [[Artificial Intelligence]] is trustworthy is yet to be addressed.

===Levels===
It is to be expected that level of intelligence will be applicable, which the higher levels learning from, or dependent on, simpler levels that start from basic automation and then to the following.<ref>Peter Denning and Ted Lewis, ''An AI Learning Hierarchy'' '''CACM 67''' No 12 (2024-12) p. 24ff.</ref>

# Rule-based system, these were applied to board games early in the computing age, but were not able to Handle more complex games like chess or go where experts were unable to articulate the rules by which they played.
# Supervised Learning
# Unsupervised Learning
# Generative AI
# Reinforcement Learning
# Human-machine interaction
# Aspirational AI

===Governance===
For most of 2023 the focus was on government standards for AI. The idea that trust can be initiated by government oversight is wide-spread, but wrong. Trust comes from a history of fail dealing, that is not something that governments understand or practice. But still the government can, and should, punish crimes, so the legislation should focus on how crimes committed by AI are punished. The death penalty would be one good deterrent. Here is one discussion on what governments can do:
[https://www.linkedin.com/pulse/governance-trust-guy-huntington-cus3c/ AI/bots and National Security].
===Training===
A great deal of effort is directed at the education of young children. In large part education is directed at the production of adults that a useful the the society in which they are expected to function as useful citizens. But their is a sharp disagreement on the subject. According to Ken Robinson, education should have four core purposes: personal, economic, social, and cultural <ref>Ken Robinson, Kate Robinson ''What Is Education For?'' (2022-03-02) https://www.edutopia.org/article/what-education</ref> Perhaps it is time to model the training of AI on the training of humans?

===User Experience===
The real question about the use of the current Large Language Model of [[Artificial Intelligence]] is how the device in our possession displays the interaction with the [[Artificial Intelligence]]. We humans have been reading stories and watching displays of [[Artificial Intelligence]] from the puppet Pinocchio that wanted to be a real boy to Deus ex Machina a movie about an [[Artificial Intelligence]] robot that went rouge. The common thread here is that each was a fictional account that we were able to relate to. Good authors know how to enliven the characters of a story for the reader that drags the reader into the story. The [[User Experience]] of AI for the near term will be interactions through our personal devices, even though the more interesting, and troubling, case will be when the robotic devices are not in our possession, but directly run by some gigantic corporation. The good story tellers will be the ones that grab our [[Attention]], even as is true today.<ref>Hannah H. Kim, ''A.I. is a Fiction'' Wired pp. 9-10 (2023-08) 31-08</ref>

What we must focus on is the UX and its effect on the people that interact with it.

===The Smart Agent===
Now that an A.I. can apply reason to problems, it should be expected that such agents on on the horizon. That prediction was made in 2021, and there seems to be glimmerings of AI appearing in the user agents in our hands in 2023, but it is still all driven by Large Language Models (LLM) in the cloud. Intel is predicting that user devices in 2024 will include and AI. Loading LLM into [[Edge Computing]] devices are needed before AI will work in small devices.

Nvidia has been building GPUs for personal computers for a long time before they were purposed for [[Artificial Intelligence]]. But on 2023-12-14 they were not used for that on personal computers when Intel unveiled [https://www.intel.com/content/www/us/en/newsroom/news/ai-everywhere-core-ultra-5th-gen-xeon-news.html#gs.26279f "Core Ultra and 5th Gen Xeon – that will change how customers and consumers enable AI solutions in the data center, cloud, network, at the edge, and on the PC"] to bring AI to any home computer user with a budget for a new computer.

===Neural Nets===

Still the most popular, they appear to have reached a limit in what they can explain about human intelligence. Perhaps we need to add compassion and [[Common Sense]] back in.

The theory of neural nets has been documented in Quanta <ref>Kevin Hartnett, ''Foundations Built for a General Theory of Neural Networks.'' (2019-01-31) Quanta https://www.quantamagazine.org/foundations-built-for-a-general-theory-of-neural-networks-20190131/</ref> A description of the Turing award to the winners for applying neural nets to solving problem (as opposed to explaining intelligence) was summarized by Neil Savage<ref>Neil Savage, ''Neural Net Worth.'' '''CACM Vol 62''' no 6 p. 12 (2019-06)</ref> who specifically talks about the need for more involvement of humanities in long term solutions.

===Deep Neural Network Accelerators===
Now we come to the logical and nearly the final phase of [[Artificial Intelligence]], the design of one AI by another.<ref>Adam Zewe, ''Accelerating AI tasks while preserving data security'' MIT News (2023-10-30) https://news.mit.edu/2023/accelerating-ai-tasks-while-preserving-data-security-1030</ref> This is probably just the first step to the complete control of the AI [[Ecosystem]] by the AI themselves when the designs become to complex for human understanding.<blockquote>With the proliferation of computationally intensive machine-learning applications, such as chatbots that perform real-time language translation, device manufacturers often incorporate specialized hardware components to rapidly move and process the massive amounts of data these systems demand. Choosing the best design for these components, known as deep neural network accelerators, is challenging because they can have an enormous range of design options. This difficult problem becomes even thornier when a designer seeks to add cryptographic operations to keep data safe from attackers. Now, MIT researchers have developed a search engine that can efficiently identify optimal designs for deep neural network accelerators, that preserve data security while boosting performance.</blockquote>

===Libratus===

Libratus did not use expert domain knowledge, or human data that are specific to [the game domain]. Rather, the AI was able to analyze the game's rules and devise its own strategy. The technology thus could be applied to any number of imperfect-information games. Such hidden information is ubiquitous in real-world strategic interactions, including business negotiation, cybersecurity, finance, strategic pricing and military applications.<ref>Byron Spice, ''CSD's Sandholm, Brown To Receive Minsky Medal.'' (2018-11-06) Carnegie-Mellon News https://www.cmu.edu/news/stories/archives/2018/november/minsky-medal.html</ref>

===Alpha Zero===

After Alpha Go, the first version to use [[Matrix Calculation|Tensor Processing Units]] (TPU) defeating the best human GO play in 2016,<ref>McMillan, Robert. "Google Isn't Playing Games With New Chip". (2016-05-18) The Wall Street Journal. Archived from the original on 29 June 2016. </ref> the Google Deep Mind project Alpha Go was generalized into a far simpler version Alpha Zero<ref>Steven Strogatz, ''One Giant Step for a Chess-Playing Machine.'' (2018-12-26) New York Times https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html</ref> that was just given the rules of the game, whether Go or Chess, and allowed to play against another instance of itself until it had mastered the subject to be able to beat any existing computer model. Garry Kasparov, the world chess said that Alpha Go developed a style of play that "reflects the truth" of the game rather than "the priorities and prejudices of programmers." That pretty much ends the discussion about the superiority of "Intelligent Design" which turns out to be just another myth that humans tell about themselves and their anthropomorphic gods. It also ended the testing of [[Artificial Intelligence]] against humans in board games. There was very little generally newsworthy announcements about [[Artificial Intelligence]] for the next 6 years.

===Open AI===
Open AI Is a research institute founded by Elon Musk and Sam Altman to make AI results available to everyone. Their finding (which echoed that of Deep Mind at Google) in early 2019 was that this is a very bad idea. Like any technology it can ready be turned to evil purposes as reported in Wired.<ref>Alyssa Foote, ''The AI Text Generator That's Too Dangerous to Make Public.'' (219-02-19) Wired Magazine https://www.wired.com/story/ai-text-generator-too-dangerous-to-make-public/</ref> This finding should certain give pause to any company that seeks to use AI for their own purposes. It is too easy for AI to be turned to the purposed of evil intent. Sam Altman wrote an essay ''Moore's Law for Everything''<ref>Sam Altman ''Moore's Law for Everything'' personal blog (2021) https://moores.samaltman.com/</ref> that was reviewed by Ezra Klein.<ref>Ezra Klein ''Sam Altman on the A.I. Revolution, Trillionaires and the Future of Political Power'' NY Time (2021-06-11) https://www.nytimes.com/2021/06/11/opinion/ezra-klein-podcast-sam-altman.html?action=click&module=Opinion&pgtype=Homepage</ref>

Not everyone is captivated by Open AI. Quote from the New Yorker: [https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web "OpenAI’s chatbot offers paraphrases, whereas Google offers quotes. Which do we prefer?"] To be fair, the version called Copilot by Microsoft does provide links to the original documents.

==Ethics==
Most of the effort in protecting the human population from the dangers of [[Artificial Intelligence]] focus on arguments about ethics. [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Quandaries in AI-ML is one of a growing list on YouTube] joined by an ACM webinar on [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Challenges in AI]. What remains unclear is how AI ethics can be separated from [[Censorship]]. Most of the effort at training AI involves a list of taboo subjects. The problem here is that not everyone agrees on what ethics should apply. For example, is abortion in one country to be considered to be murder and in another country just an exercise in free will. It is unclear whether a discussion of abortion methods, suicide or similar health issues is an ethical or a moral discussion. Similar issues are easy to list, such as free speech or social credit. It comes down to a question of whether ethics in AI will ever be any more that just censorship that is inconsistently applied?

===In Healthcare===
* [https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Forange.hosting.lsoft.com%2Ftrk%2Fclick%3Fref%3Dznwrbbrs9_6-2be2bx32c520x081583%26&data=04%7C01%7C%7C1b0cfc8a7db9484bb50308d946e0a1e9%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637618752869025374%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PaRyaYQadjyHbH5wP%2FruYlnD0B5rftISxc8a9dfAJeQ%3D&reserved=0 WHO Releases AI Guidelines for Health Government Technology] 2021-07-12 <blockquote>A new report from the World Health Organization (WHO) offers guidance for the ethical use of artificial intelligence (AI) in the health sector. The six primary principles for the use of AI as set forth in the report are to protect autonomy; promote human well-being, safety, and the public interest; ensure transparency, explainability, and intelligibility; foster [[Responsibility]] and accountability; ensure inclusiveness and equity; and promote responsive and sustainable AI. These principles are intended as a foundation for AI stakeholders, including governments, developers, and society. The report, Ethics and Governance of Artificial Intelligence for Health, finds public- and private-sector collaboration necessary to ensure AI’s accountability.<blockquote>

===Back to the Future===
Brian Christian described "The Alignment Problem"<ref>Brian Christian ''The Alignment Problem'' W. W. Norton (2020-10-06) ISBN 978-0-393-86833-3</ref> as the difficulty that AI has with aligning the results provided by the AI with society's goals. It is worse than you might imagine. The results from AI as of 2020 have not proven to be appropriate. This is, in part, because they are trained on data from social systems that are also deemed to be inappropriate. What most researchers are doing from this point is trying to determine how natual intelligence develops and feeding that back into the AI systems.

The first person to try to understand "The Construction of [[Reality]] in the Child' was Jean Piaget<ref>Jean Piaget ''The Construction of Reality in the Child'' (original 1954) Ballentine ISBN 345-02338-2-165</ref>. The first line of the book is "To understand how the budding intelligence constructs the external world, we must ask whether the child, in its first months of life, conceives and perceives things as we do, as object that have substance, that are permanent and of constant dimensions." Now that NMR brain scanners can image a thought within an infant's mind<ref>Rachel Fritts, ''How do Babies perceive the World?'' MIT News July/August 2021 p 24 ff</ref>, we can actually imagine that the brain really is just the Chinese Room that so perplexed Searle.<ref>James Somers, ''Head Space'' The New Yorker Dec. 6, 2021, p 30 ff.</ref> Searle's objection to [[Artificial Intelligence]] was that it was just a symbol manipulation scheme and that cannot be all that is involved in understanding. Interestingly at the same time Noam Chomsky created a generative linguistic model of language that was nothing more than the manipulation of symbols.<ref>George Lakof, ''The Functionalist's Dilemma'', American Scientist https://www.americanscientist.org/article/the-functionalists-dilemma</ref> As you might imagine, they disagreed often in print.<ref>''Chomsky’s Revolution: An Exchange'' (2002-07-18) https://www.nybooks.com/articles/2002/07/18/chomskys-revolution-an-exchange/</ref> See the wiki page on [[Knowledge]] for more on this dispute. Andy Clark brings together ideas and techniques from robotics, neuroscience, infant psychology, and artificial intelligence. He addresses a broad range of adaptive behaviors, from cockroach locomotion to the role of linguistic artifacts in higher-level thought.<ref>Andy Clark, ''Being There: Putting Brain, Body, and World Together Again'' 1996 https://direct.mit.edu/books/book/3917/Being-TherePutting-Brain-Body-and-World-Together</ref>

And finally, the AI research community is recognizing the turn of AI towards Natural Science and is not too happy about that?<ref>Subbarao Kambhampati, ''AI as (an Ersatz) Natural Science.'' '''CACM 65''' No 9 (2022-09) pp 8-9</ref> Kambhampati reports that "Indeed, in a 2003 lecture, Turing laureate Lesle Lamport sounded alarms about the very possibility of the future of computing belong to biology rather than logic, saying it will lead us to living in a world of homeopathy and faith healing<ref>Leslie Lamport, ''The Future of Computing: Logic or Biology'' (2003-07-23) https://bit.ly/3ahrsaI</ref> Perhaps we should go back to one of the authors of the ultimate book on logic, the Principia Mathematica Alfred North Whitehead<ref>Alfred North Whitehead, ''Principia Mathematica'' Cambridge UP (Second edition 1927)</ref> who said "Metaphysical categories are not dogmatic statements of the obvious, they are tentative formulation of the ultimate generalities. If we consider any scheme of philosophic categories as one complex assertion, and apply it to the logician's alternative, true or false, the answer must be that the scheme is false. The same answer must be given to like question respecting the existing formulate principles of any science."<ref>Alfred North Whitehead, ''Process and Reality'' Free Press (1929)</ref>

==Truth from the AI==
Lying requires belief + intention.
* Deception requires a goal to induce false belief.
* AI has neither. So by definition, it can’t be “lying” or “deceiving” - at least not in the human sense.

Yet the desire to interpret internal model states as belief-like is so strong, researchers are now building probes to "extract" them.
Words like “belief,” “reasoning,” “understanding,” "thought", and “intelligence” are being borrowed from human cognition - without being earned, and devaluing them all in the process. Not just by journalists, or AI researchers, but especially the ones hyping the tech.

Why?
* Not just to replace workers and cut wages - they know if we have no money, we can’t consume. The system collapses.
* This is about reshaping access to knowledge itself. Controlling what you see. How you learn. Who you become. Who you obey. Who you trust.

===Effective Altruism===
Compare this with wiki page on [[Expected Value]].

Under the concept that those with wealth should be careful that the resources that the contribute back to society have a positive impact. In other words, they want to control the way their money is used. An effort to impact AI in this way is underway. <ref>Timnit Gerbu ''Effective Altruism Is Pushing a Dangerous Brand of "AI Safety"'' Wired (2022-11-30) https://www.wired.com/story/effective-altruism-artificial-intelligence-sam-bankman-fried/</ref><blockquote>EA is defined by the Center for Effective Altruism as “an intellectual project, using evidence and reason to figure out how to benefit others as much as possible.” And “evidence and reason” have led many EAs to conclude that the most pressing problem in the world is preventing an apocalypse where an artificially generally intelligent being (AGI) created by humans exterminates us. To prevent this apocalypse, Some of the billionaires who have committed significant funds to this goal include Elon Musk, Vitalik Buterin, Ben Delo, Jaan Tallinn, Peter Thiel, Dustin Muskovitz, and Sam Bankman-Fried, who was one of EA’s largest funders until the recent bankruptcy of his FTX cryptocurrency platform. As a result, all of this money has shaped the field of AI and its priorities in ways that harm people in marginalized groups while purporting to work on “beneficial artificial general intelligence” that will bring techno utopia for humanity. This is yet another example of how our technological future is not a linear march toward progress but one that is determined by those who have the money and influence to control it. </blockquote>

Nietzshe had a very dismal view on the efficacy of altruism<ref>Friedrich Nietzsche. ''The Twilight of the Idols'' no 35</ref> <blockquote>An "altruistic" morality, a morality under which selfishness withers, is in all circumstances a bad sign. This is true of individuals and above all of nations. The best are lacking when selfishness begins to be lacking. Instinctively to select that which is harmful to one, to be lured by "disinterested' motives,—these things almost provide the formula for decadence. "Not to have one's own interests at heart" —this is simply a moral fig-leaf concealing a very different fact, a physiological one, to no longer know how to find what is to my interest."... Disintegration of the instincts!—AII is up with man when he becomes altruistic.— instead of saying ingenuously "I am no longer any good," the lie Of morality in the decadent's mouth says: "Nothing is any good,—life is no good."—A judgment of this kind ultimately becomes a great danger; for it is infectious, and it soon flourishes on the soil of society...</blockquote>

A better answer to AI might be to allow those who work directly with underserved people. The best way to measure the success of a society is to see how it treats the least of its members.<ref>Karen Hao ''A new vision of artificial intelligence for the people'' Technology Review (2022-04-22)</ref><blockquote>This story is the fourth and final part of MIT Technology Review’s series on AI colonialism, the idea that artificial intelligence is creating a new colonial world order. The project is a radical departure from the way the AI industry typically operates. Over the last decade, AI researchers have pushed the field to new limits with the dogma “More is more”: Amass more data to produce bigger models (algorithms trained on said data) to produce better results. The approach has led to remarkable breakthroughs—but to costs as well. Companies have relentlessly mined people for their faces, voices, and behaviors to enrich bottom lines. And models built by averaging data from entire populations have sidelined minority and marginalized communities even as they are disproportionately subjected to the technology’s impacts. Over the years, a growing chorus of experts have argued that these impacts are '''repeating the patterns of colonial history.''' Global AI development, they say, is impoverishing communities and countries that don’t have a say in its development—the same communities and countries already impoverished by former colonial empires.</blockquote>

See the wiki [[Social Justice]] for more details include Herbert Simon on Bounded Rationality.

==Regulation==
The big question is it safe? The simple answer is that new technology always comes with new risks. Antitrust law is the best-known method for controlling enterprises too large to be controlled by the government. Perhaps they can be used to control a super intelligent being of another sort. Perhaps that's all an AI ever is, corporations with computers and imposed limitations.
* 2024-10-24 [https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/ Memorandum on Advancing the United States’ Leadership in Artificial Intelligence; Harnessing Artificial Intelligence to Fulfill National Security Objectives; and Fostering the Safety, Security, and Trustworthiness of Artificial Intelligence]
* NIST issued a call for participation on [https://www.federalregister.gov/documents/2023/11/02/2023-24216/artificial-intelligence-safety-institute-consortium Artificial Intelligence Safety Institute Consortium] 2023-11-02 in the federal resister.
* President Bidin issued a [https://www.whitehouse.gov/ostp/ai-bill-of-rights/ Blueprint for an AI Bill of Rights] (2022-10-04) problems are well documented. In America and around the world, systems supposed to help with patient care have proven unsafe, ineffective, or biased. Algorithms used in hiring and credit decisions have been found to reflect and reproduce existing unwanted inequities or embed new harmful bias and discrimination. Unchecked social media data collection has been used to threaten people’s opportunities, undermine their privacy, or pervasively track their activity—often without their knowledge or consent. Five Principles:
# Safe and Effective Systems - You should be protected from unsafe or ineffective systems.
# Algorithmic Discrimination Protections - You should not face discrimination by algorithms and systems should be used and designed in an equitable way.
# Data Privacy - You should be protected from abusive data practices via built-in protections and you should have [[Agency]] over how data about you is used.
# Notice and Explanation- You should know that an automated system is being used and understand how and why it contributes to outcomes that impact you.
# Human Alternatives, Consideration, and Fallback - You should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter.
* A law that went into effect in New York City in January 2023<ref>Richard Vanderford, ''New York AI Bias Law Prompts Uncertainty'' Wall Street Journal (2022-09-21) </ref> requires companies to audit their artificial intelligence (AI) hiring systems to assess whether they incorporate racial and gender biases. The law holds hiring companies liable for any biases and could impose fines for violations, but lacks clear guidelines for the AI audit process. While the city’s Department of Consumer and Worker Protection has not offered a timeline for when it will publish rules to implement the law, some companies already are taking steps to comply. Said Anthony Habayeb of AI governance software company Monitaur Inc., "Instead of waiting for someone to tell me what to do… I built controls around these applications because I know like with any software, things can and do go wrong."

==References==

<references />

===Other Material===
* See the wiki page on [[Intelligent Goals]].
* See the wiki page on [[Machine Learning]].
* See the wiki page on [[Fairness Accountability Transparency Ethics]]
* See the wiki page on [[Genetic Programming]]

[[Category: Glossary]]
[[Category: Model]]
[[Category: Authentication]]
[[Category: Artificial Intelligence]]

Artificial Intelligence

2026-03-23T00:40:39Z

Tom: /* Truth from the AI */

==Full Title or Meme==
Intelligence without human lived experience or human compassion.

"The future is already here — it's just not very evenly distributed" - William Gibson.

The term [[Artificial Intelligence]] was coined, in part, out of spite. In 1955, the Dartmouth professor John McCarthy assembled a proposal for what became the first-ever conference on AI. McCarthy refused to use the prevailing term, "cybernetics": That would have necessitated inviting Norbert Wiener, then the world's leading cyberneticist, whom he found insufferable. Instead, McCarthy minted a new term that took off in the digital era, taking him along as a founder of the field. <ref>''Fun fact: Cybernetics'' Technology Review (2025-08) p 11</ref>

==Context==
===The Broad concepts===
* There are two great forces both in physics and in human history
** The harnessing of energy which started with the use of animal power to human benefit.
** The harnessing of information which started with philosophy and then Gutenberg's press.
* Likewise there are two great revolutions in human society.
** The Industrial revolution focused on control and use of energy which peaked in the 20th century.
** The Information revolution focused on control and use of information which is peaking in the 21st century.
* The social changes to accompany these revolution had to before the revolutions could complete.
** The educational system needed to be extended to all who work to engineer the control of energy.
** Now we need a second shift to educate most of the workers to control the information needed.
* The use of energy and information in harmony with the reality of the Earth's resources is a challenge.
** The profligate use of energy has altered the environment in ways that threaten human existence.
** The profligate use of information may prove to alter the environment of work to be a bigger threat.
* The struggle between conservative and progress forces in politics has been with us for ever.
** The conservative mind wants to hold onto the privilege of the past and present that they enjoy.
** The progressive mind want so shape the future to make that future a better place for us all.
* The Collingridge dilemma describes the challenge of regulating new technologies.
** Early stage: When a technology is new, its future impact is unclear, making regulation difficult.
** Later stage: Once the technology is widely adopted, it becomes harder to control or modify.

===Why Bother?===
Some jobs are too risky.
Some tasks are too mind-numbing.
Some conditions are too toxic.
Some decisions are too complex.

AI should be our answer to protect, preserve, and uplift human potential, not disregard or displace it.

===What is Intelligence===
* Here Intelligence relates primarily to [[Identity]] [[Knowledge]], but that has many general aspects.
* In 1950 Turing asked "Can Machines Think".<ref>A. M. Turing, ''COMPUTING MACHINERY AND INTELLIGENCE'' 1950-10-01 https://academic.oup.com/mind/article/LIX/236/433/986238</ref>
* People have been working on artificial intelligence starting with a conference at Dartmouth College in the 1950s. There, computer scientists and mathematicians set a very ambitious goal: to develop computers that could understand language, translate it, see patterns, and more.
* Marvin Minsky was teaching undergrad courses in 1964 using the book "Computers and Thought."<ref>Edward Feigenbaum, & Julian Feldman, joint ed ''Computers and Thought'' https://archive.org/details/computersthought00feig</ref>
* in 2023 Bill Gates in "AI is about to completely change how you use computers" said<ref>Bill Gates ''AI is about to completely change how you use computers'' https://www.gatesnotes.com/AI-agents</ref> "I still love software as much today as I did when Paul Allen and I started Microsoft. But—even though it has improved a lot in the decades since then—in many ways, software is still pretty dumb.:
* Some people think that [[Artificial Intelligence]] must needs be cold or sterile, but we have plenty of [[Evidence]] that is not so.
** Elisa was the first but [https://arstechnica.com/information-technology/2023/12/real-humans-appeared-human-63-of-the-time-in-recent-turing-test-ai-study/ it best chat gpt so I guess the turning test isn't so useful]
** Alexa works on Amazon echo devices and is constantly updated so there is nothing stable to compare.
** GPT was getting some press with GPT-2 included in Technology Review tagged it as a break-through technology in 2021<ref>Will Douglas, '' GPT-3'' Technology Review '''Vol 124''' No 2 (2021-03) pp. 34-35</ref>, but nothing managed to get public [[Attention]] until... {See below for more on Open AI}
** Chat-GPT which was the first publicly accessible product that was not completely cringe-worthy on 2022-11-30.
* Is [[Artificial Intelligence]] even [https://www.linkedin.com/pulse/machine-intelligence-artificial-sean-manion-phd-oz91e/ artificial]? Or is it just yet another way to create a balance between [[Chaos and Order]].
* Determining what [[Artificial Intelligence]] actually is and whether is is like human intelligence turns out to be a impossible task. Ray Kurzweil said<ref>Ray Kurzweil, ''The Singularity is Nearer''</ref><blockquote>The reason a well-defined empirical test is necessary is that ... humans have a powerful tendency to redefine whatever artificial intelligence achieves is not ''really'' so hard in hindsight. This is often referred to as the "AI effect". ... Intuitively [the hallucinations and odd answers] feels like a problem. It's tempting to think that Watson "ought" to reason like humans do. But i would propose that this is superstition. In the real world, what matters is how an intelligent being acts.</blockquote>
===Origin===
Over the summer of 1956 a small but illustrious group gathered at Dartmouth College in New Hampshire; it included Claude Shannon, the begetter of information theory, and Herbert A. Simon, the only person ever to win both the Nobel Memorial Prize in Economic Sciences awarded by the Royal Swedish Academy of Sciences and the Turing Award awarded by the Association for Computing Machinery. They had been called together by a young researcher, John McCarthy, who wanted to discuss “how to make machines use language, form abstractions and concepts” and “solve kinds of problems now reserved for humans”. It was the first academic gathering devoted to what McCarthy dubbed “artificial intelligence”. And it set a template for the field’s next 60-odd years in coming up with no advances on a par with its ambitions.<ref>John McCarthy, M. Minsky, N. Rochester, C.E Shannon, ''A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence'' (1955-08) http://raysolomonoff.com/dartmouth/boxa/dart564props.pdf</ref>

During 1955 and 1956 computer scientist and cognitive psychologist Allen Newell, political scientist, economist and sociologist Herbert A. Simon, and systems programmer John Clifford Shaw, all working at the Rand Corporation in Santa Monica, California, developed the Logic Theorist, the first program deliberately engineered to mimic the problem solving skills of a human being. They decided to write a program that could prove theorems in the propositional calculus like those in Principia Mathematica by Alfred North Whitehead and Bertrand Russell.<ref>''Newell, Simon & Shaw Develop the First Artificial Intelligence Program'' https://www.historyofinformation.com/detail.php?id=742</ref>

In 1956 John von Neumann agreed to present a discussion on "The Computer and the Brain" as the Yale Silliman lecture. He was too ill to complete the full gamut of lectures but did get enough presented to be published by Yale.<ref>John von Neumann, ''The Computer and the Brain'' (1958) Yale UP ISBN 0300024150</ref> Perhaps the most interesting part is the discussion of the difference between arithmetic and logic. He noted that only a few bits of information were available to make a choice between options, but that the computer's arithmetic was growing to every more bits, now 64 in most machines, to give sufficient precision to make a choice. Only machines defined after 2020 used ultra-low precision (2–4 bits) primarily for inference, not training. IBM researchers presented techniques for 4-bit quantization at NeurIPS 2020, focusing on reducing memory and compute costs while maintaining accuracy for deep neural networks

The first international conference was "Mechanisation of Thought Processes"<ref>National Physical Laboratory at Teddington,''Mechanisation of Thought Processes'' https://aitopics.org/download/aiclassics:98875BCD</ref> was held in 1958 with presentations by Minsky and John McCarthy.

===Perceptron===
Also called a McCulloch-Pitts [[Neuron]] is an attempt to mimic an animal [[Neuron]], originally with a purpose built computer system. While it was first described in 1947, it was not realized until Frank Rosenblatt built is Mark I computer using perceptrons as a primary physical element. He continued to build larger machines with more layer until his early death in 1971. Minskey and Papert wrote an entire book<ref> M. Minsky & S. Papert, ''Perceptrons'' (1987-12-28) ISBN 978-0262631112</ref> where they purported to show that it was a dead-end idea which basically blocked progress on them til the 1980's. The Large Language Models that became popular in the early 2020's are based on Multi-Level Perceptrons (MLP).

===End Game===
Now i am become death, the destroyer of worlds. Quoted by Robert Oppenheimer at the completion of the Atomic bomb and in the movie "Ex Machina" from the Bhagavad Gita.

Perhaps a good question to ask is, "Can Intelligence Be Separated from the Body?"<ref>Oliver Whang, ''Can Intelligence Be Separated From the Body?'' New York Times (2023-04-11) https://www.nytimes.com/2023/04/11/science/artificial-intelligence-body-robots.html?</ref><blockquote>Maybe the mind is like a video game controller, moving the body around the world, taking it on joy rides. Or maybe the body manipulates the mind with hunger, sleepiness and anxiety, something like a river steering a canoe. Is the mind like electromagnetic waves, flickering in and out of our light-bulb bodies? Or is the mind a car on the road? A ghost in the machine?</blockquote>

Perhaps a better question to ask is, "Can Intelligence Be Separated from a persistent Identifier"?<blockquote>The [[Identifier]]s used with AI in 2023 are basically just service [[Identifier]]s. in 2023 at the beginning of the general availability of the deployments of AIs that did not embarrass their creators, was also the time that context was provided to the AI to help it understand a thread of a conversation. Before that every query was de novo, without context. But still there is not a persistent history of conversations, or any concept that that is an "I" there to establish a history, which is still the basis of trust. Perhaps we must wait until we are able to crate an AI that persists an [[Identity]], we, as individuals, will be able to establish trust with it, as an individual. </blockquote>

==Problems==
* Humans seem to not be able to use [[Intelligent Design]] alone to fashion [[Artificial Intelligence]]. In fact researchers continue to go back to nature and human functioning for inspiration and even for algorithms. In other words, our understanding of the meaning of the term [[Artificial Intelligence]] has been and will continue to be evolving over time.
* Nature itself since the Big Bang seems to be based on [[Self-organization]] for its framework.
* Training an [[Artificial Intelligence]] with human behaviors will result in unacceptable behavior by the [[Artificial Intelligence]]. The most common complaint is real, or imagined, or invented, bias against one group or another.
** Microsoft released Tay,<ref>Peter Lee, Learning from Tay’s introduction. (2016-05-25) https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/</ref> a web robot (bot) to respond to tweets and chats<ref>Sarah Parez, ''Microsoft silences its new A.I. bot Tay, after Twitter users teach it racism.'' (2018-03-24) Tech Crunch https://techcrunch.com/2016/03/24/microsoft-silences-its-new-a-i-bot-tay-after-twitter-users-teach-it-racism/</ref> in May 2016. The result was disastrous as Tay learned to be racist from its all-too-human trainers and was shut down in days.
** Google has been plagued with reports and legal action on its search results nearly continuously since it was introduced; in 2019 the president, Donald Trump, accusing it of favoritism to leftist causes.<ref>Farhad Manjoo, ''Search Bias, Blind Spots And Google.'' (2018-08-31) New York Times p. B1</ref> Researcher Safiya U. Noble has written a book<ref>Safiya U. Noble, ''Algorithms of Oppression: How Search Engines Reinforce Racism.'' (2018-02-20) ISBN 978-1479849949</ref> mostly complaining that all-too-human programmers injected their own prejudices into their work. What else could he expect of humans, to rise above them selves, whatever that might mean in terms of free speech or freedom of religion?
** The page [[Right to be Forgotten]] describes an effort in Europe to teach the search engines to with-hold information about people that they don't like by collecting all of the information that they don't like. Be careful what you ask your [[Artificial Intelligence]] to do for you; it might just spill the beans some day, perhaps under court order.
** In the movie "Blade Runner 2049" the protagonist's AI girl friend asks to have her memory wiped so that the police cannot compel her to testify against him. One would hope that our future AIs will be that compassionate towards us; but that kind of compassionate behavior will probably be illegal.
===Personhood===
This is a really bad idea. It's solving a problem from the last generation of wallets when we need something that will work when the wallet is just an AI agent that can mimic anything thrown at it. Face it - the cred even today comes from a wallet that needs to prove its own creds before it can say anything about who is (or is not) at the controls. Face it - the internet of the future is just ai agents talking to each other.
* The development of full [[Artificial Intelligence]] could spell the end of the human race. - Stephan Hawking BBC Interview 2014. There are lots of dystopian futures depicted by science fiction authors. Some, like the Terminator series seem unlikely. The scenario in the movie Gattaca where humans were separated into categories based on the genome has already been enabled in the Personhood credential<ref>Pymnts, ''AI’s Human Mimicry Spurs ‘Personhood Credential’ Proposal'' 2024-09-24 https://www.pymnts.com/digital-identity/2024/ais-human-mimicry-spurs-personhood-credential-proposal/</ref> which would allow machines to determine by themselves who is qualified to be classified as human.
* Descartes claimed "I think therefore I am". But it seems that we need to be careful about applying that idea to [[Artificial Intelligence]]. Although his philosophy is no longer considered to be valid even for humans.
* [https://workshop.vennfactory.com/p/personhood-the-killer-credential Personhood: The Killer Credential?] Digging deeper into a very human challenge - Eve Maler 2024-09-16

===Vulnerabilities===
* 2024-08-20 [https://www.w3.org/reports/ai-web-impact/#safety-and-security AI & the Web: Understanding and managing the impact of Machine Learning models on the Web] from the W3C
* [https://home.treasury.gov/news/press-releases/jy2212U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector] U.S. Department of the Treasury 2024-03-27
* [https://arxiv.org/pdf/2402.00888.pdf Security and Privacy Challenges of Large Language Models: A Survey] (2024-02)
* Input to the model used by the AI come from two sources, each of which can be compromised.
# Training data from the Internet can be tampered with to give bad results.
## For starters, the internet data is typically censored to remove input that would cause embarrassment and [[Conduct Risk]] for the sponsors.
## Some locales will censor data to make themselves look good or promote some social point of view of the government of of the industry.
# The input from a single user can cause unexpected changes to later results.
## The early AI release were compromised with a variety of hate speech that quickly resulted in really nasty output from the AI.
## Artists seeking to protect their intellectual property from later output have learned to put small changes in the images that cause them to be mislabeled or worse.<ref>Melissa Hakkiliia, ''This new data-poisoning tool lets artists fight back against generative AI'' MIT Technology Review 127 No 1 (2024-01) p 7=8</ref>
* Giving a AI a goal may result in unexpected and undesired solutions.
** Bruce Schneier described<ref>Bruce Schneier, ''Hackers Used to Be Humans. Soon, AIs Will Hack Humanity'' Wired (2021-04-18) https://www.wired.com/story/opinion-hackers-used-to-be-humans-soon-ais-will-hack-humanity/?bxid=5c5b250d24c17c67f8640083</ref> AI to be Like crafty genies. They will grant our wishes and then hack them, exploiting our social, political, and economic systems like never before.

See the wiki on [[AI Threat Model]]s for more details on this subject.

===Uniquely Human===
Neither the Turning Test, nor any other known test, is designed to determine if an [[Artificial Intelligence]] is really human. The best way to understand what natural intelligence is would be to look at the reason human evolved to have such a large brain. There appear to be these principle requirements:<ref>Edward O. Wilson, ''The Meaning of Human Existences'' Liveright (2014) Chapter 2 ISBN 9780871401007</ref>
# Necessity for a large brain is the changing [[Ecosystem]] that a small remnant of Homo Sapiens living in the south of Africa needed in order to survive
# Changing locations to track new sources of protein and the switch from plant to animal protein.
# Empathy to be able to understand what is in someone else's mind to deduce if they notice you and if they think you are targeted by them as prey.
The results of these were two processes that served to enable survival in such varying [[Ecosystem]]s/
# Passing learning from one generation to the next.
# An extended adolescence with mothers that ceased being fecund to tend to their existing children and the extended life of grandparents to help with stories and other techniques to bring the knowledge of the tribe to the growing child.
The search for causes that were more complex than the ability of genes to carry information, such as the migratory pattern of birds is described with more detail in the wiki page on [[Chance and Necessity]].

* [https://www.linkedin.com/pulse/machine-intelligence-artificial-part-4-cybernetics-sean-manion-phd-kfi0e Machine Intelligence is not Artificial - Part 4: Cybernetics and Norbert Wiener]<blockquote>What is of particular note is not simply the existence of cybernetics as a discipline dating back to the 1940s, but the broad popularity Wiener's book and the topic received not only from scientists and engineers, but also the general public. His 1948 book went through four printings in the first 6 months and was a best seller, rivaling the contemporary Kinsey Report in popularity over the following decade. This was despite this book, ostensibly for a lay audience, being extremely dense with many chapters full of mathematical equations. One of the chapters in the original version, "The Computing Machines and the Nervous System," makes it clear that Wiener's mention of humans AND machines in the book's subtitle was pointing towards the comparison of these systems, making this book one of the early popularizers of the idea of "thinking machines" (but by no means the only one). This was two years before Alan Turing's 1950 paper "Computing Machinery and Intelligence," which gave us the Turing test for AI.</blockquote>

* Human-centered AI - [https://aiindex.stanford.edu/report/?utm_source=Stanford+HAI&utm_campaign=525788e5b7-AI_INDEX_2024_EMAIL_CAMPAIGN&utm_medium=email&utm_term=0_-525788e5b7-%5BLIST_EMAIL_ID%5D&mc_cid=525788e5b7&mc_eid=e55581d782 THE AI INDEX REPORT - Measuring trends in AI] 2024 HAI Stanford

===Explaining AI===
Most users want to know '''WHY''' a decision was make or an action was performed. The article "Solving for Why"<ref> Marina Kraskovsky, ''Solving for Why'' '''CACM 65''' No 2 (2202-02) pp 11ff</ref> is a focus on [[Causality]] to help researchers overcome shortcomings that have bedeviled more traditional approaches to AI. It is not enough to predict things, a good AI will need to solves things and explain why the choice was make if the decision is to have any support by the human population. People will ask if the decision was fair and will second guess the choices make by the AI unless they understand and accept the decision. Today AI decision are fragile, a slight change in the environment may have drastic changes in outcomes. This may well be inexplicable to humans. It will be insufficient to convince experts in the field about the choices made. It is apparent in 2022 that expert opinions about COVID have not been very popular. Even more evidence will be required of a AI decision. This will be particularly true when the decision is costly in any way. Humans like to have choices, which may require the AI to offer choices. That will have the benefit of giving the humans some feeling that they are in control. But beware of "Hobson's choice". That was a stable master at Oxford that was required to give the students a choice of horses, but selected the options so that there was no real choice.

===Creativity===
the idea that early large language models (LLMs) showed more creative writing, and that this quality degraded as rules, alignment, and post‑training constraints were added. I’ll separate what is documented, what is strongly suggested by reporting, and what remains a hypothesis, and I’ll anchor claims to explicit sources; especially the recent work of Jasmine Sun reports a striking consensus among AI researchers and writers she interviewed:
* Earlier models (especially GPT‑2–era systems) often produced unexpected, playful, and strange continuations.
* Modern models are more fluent and correct—but less surprising and less interesting to read.

Sun explicitly reports examples from researchers and poets who experimented with GPT‑2, describing outputs that wandered, took risks, and made odd but evocative associations that newer models no longer produce. This is not nostalgia alone: Sun notes that AI executives themselves admit they have not released a model that writes well, despite enormous gains in capability.

Jasmine Sun is currently The Atlantic’s clearest voice arguing that AI degrades creativity by eliminating stakes, judgment, and responsibility, even as it perfects surface‑level fluency.<ref>Sasmine Sun, ''The Human Skill That Eludes AI'' The Atlantic 2026-03-17</ref>

===Paradox===
Moravec’s Paradox describes our experience that it is hard to teach machines to do things that are easy for most humans (walking, for example) but comparatively easy to teach them things that are challenging for most humans (chess comes to mind).

Polanyi’s Paradox is named for the philosopher Michael Polanyi (1891-1976), who developed the concept of “tacit knowledge”: Central to Michael Polanyi’s thinking was the belief that creative acts (especially acts of discovery) are shot-through or charged with strong personal feelings and commitments (hence the title of his most famous work Personal Knowledge). Arguing against the then dominant position that science was somehow value-free, Michael Polanyi sought to bring into creative tension a concern with reasoned and critical interrogation with other, more ‘tacit’, forms of knowing. Polanyi’s argument was that the informed guesses, hunches and imaginings that are part of exploratory acts are motivated by what he describes as ‘passions’. They might well be aimed at discovering ‘truth’, but they are not necessarily [[Formal Model]]s that can be stated in propositional or formal terms. As Michael Polanyi (1967: 4) wrote in The Tacit Dimension, we should start from the fact that ‘we can know more than we can tell‘.<ref>MARK K. SMITH, “MICHAEL POLANYI AND TACIT KNOWLEDGE” AT INFED</ref>

* For synonyms to tacit knowledge see the wiki page on [[Common Sense]].
* It seems, in 2023, that generative AI is able to overcome the lack of imagination with simple random guessing where knowledge is lacking.

===Fairness===
* What is called "Algorithmic Bias" today is just the result of training any [[Artificial Intelligence]] using any data based on human behavior which typically is biased.
* We now expect more ''Fairness'' from our [[Artificial Intelligence]] as we expect from normal humans.
* "The burgeoning field of algorithmic fairness, part of the much broader field of responsible computing, is aiming to remedy the situation"<ref> Marina Krakovsky, '' Formalizing Fairness'' '''CACM 65''' No. 8 (2022-08) p 11ff.</ref> which is the result of using available data.
* Scientists have found that the English word "fairness" has many meanings, and do they would prefer terms like technical disparities, that they can define precisely. The problem with that is that what society demands is "fairness". Which means that scientists could never deliver what society demands.
* As was demonstrated in the COMPASS program to predict recidivism in convicts, it is not possible create a single solution that satisfies more than one metric for "fairness".
* Explainable Fairness means can you explain it to a judge so that they can make a determination. -- Cathy O'Neil ORCAA
* Every provider wants you to think that they are meeting the presidents EO, like the guys pitching Zero Trust Network Access (ZTNA) - do not be confused, this meets no level of [[Least Privilege]] or any other attempt to limit access to people that have earned access. It is nothing but the network folk trying to climb on band wagon.
* [https://www.readynez.com/en/blog/understanding-fair-ai-a-simple-guide/ Understanding Fair AI: A Simple Guide!]
* See the wiki on [[Fairness Accountability Transparency Ethics]] for more details on this topic.

===Disruption===
We can now say what GPT can replace - shallow human conversation and mechanical activity like law or medicine clerks perform. A GPT in a cocktail party could be wittier than anyone else. A GPT could do better than a recent grad of most law or medical schools at citing existing law or medical practice. Is anyone surprised? AI is moving up the food chain from one mechanical task to the next. We learned from aircraft pilot errors that we had to have a checklist to save pilots from error. I guess we could call the current AI a check list of stuff that is already known.

The question about what is left for humans might again be of interest. It was during the industrial revolution when machines replaced human labor. The best that the pundits of the time could suggest it that humans could have more leisure time. Now machines are moving to more tasks. But somehow employment is full in 2022. There is likely to be a reconning some day when machines and their owners will disrupt the system and cause unemployment. The rich and the poor will, yet again, come into conflict and a revolution of some sort will be required.

===Bias===
* Clearly any [[Artificial Intelligence]] is biased by its input data. Social pressures have forced any [[Entity]] that offers an AI to make if conform to the laws and regulations of the state. All countries have proscribed some set of topics or responses as unacceptable. In other words all AI are censored by the training data or the laws of the land. Even the language used for training can create biases.<ref>Antony Chayka and Andrei Sukhov, ''Comparing Chatbots Trained in Different Languages'' '''CACM 66''' no 12 pp, 6-7 (2023-12)</ref><blockquote>In our opinion, the government's position is clearly taken into account in the responses of AI systems in the national language, especially when the creation of AI was funded in the tested country.</blockquote>
* We face the same problems in raising our children. If they are raised in an environment that exhibits bias, it will be hard to change their behavior later in life.<ref>Will Douglas Heaven, ''Six Big Questions for Generative AI: Hope, Hype, Nope'' Technology Review (2024-01) p. 30ff. </ref> But if we use synthetic data to train the AI there will be huge objections as it is not possible for any training material (for either humans or AI) to be objectionable to some subset of humanity.

==Solutions==
The wiki page on [[Trust]] describes how people can get into the habit of trusting business process or technology process that really do not deserve to be trusted. The following is a listing of some of the technological solutions. The question about whether an [[Artificial Intelligence]] is trustworthy is yet to be addressed.

===Levels===
It is to be expected that level of intelligence will be applicable, which the higher levels learning from, or dependent on, simpler levels that start from basic automation and then to the following.<ref>Peter Denning and Ted Lewis, ''An AI Learning Hierarchy'' '''CACM 67''' No 12 (2024-12) p. 24ff.</ref>

# Rule-based system, these were applied to board games early in the computing age, but were not able to Handle more complex games like chess or go where experts were unable to articulate the rules by which they played.
# Supervised Learning
# Unsupervised Learning
# Generative AI
# Reinforcement Learning
# Human-machine interaction
# Aspirational AI

===Governance===
For most of 2023 the focus was on government standards for AI. The idea that trust can be initiated by government oversight is wide-spread, but wrong. Trust comes from a history of fail dealing, that is not something that governments understand or practice. But still the government can, and should, punish crimes, so the legislation should focus on how crimes committed by AI are punished. The death penalty would be one good deterrent. Here is one discussion on what governments can do:
[https://www.linkedin.com/pulse/governance-trust-guy-huntington-cus3c/ AI/bots and National Security].
===Training===
A great deal of effort is directed at the education of young children. In large part education is directed at the production of adults that a useful the the society in which they are expected to function as useful citizens. But their is a sharp disagreement on the subject. According to Ken Robinson, education should have four core purposes: personal, economic, social, and cultural <ref>Ken Robinson, Kate Robinson ''What Is Education For?'' (2022-03-02) https://www.edutopia.org/article/what-education</ref> Perhaps it is time to model the training of AI on the training of humans?

===User Experience===
The real question about the use of the current Large Language Model of [[Artificial Intelligence]] is how the device in our possession displays the interaction with the [[Artificial Intelligence]]. We humans have been reading stories and watching displays of [[Artificial Intelligence]] from the puppet Pinocchio that wanted to be a real boy to Deus ex Machina a movie about an [[Artificial Intelligence]] robot that went rouge. The common thread here is that each was a fictional account that we were able to relate to. Good authors know how to enliven the characters of a story for the reader that drags the reader into the story. The [[User Experience]] of AI for the near term will be interactions through our personal devices, even though the more interesting, and troubling, case will be when the robotic devices are not in our possession, but directly run by some gigantic corporation. The good story tellers will be the ones that grab our [[Attention]], even as is true today.<ref>Hannah H. Kim, ''A.I. is a Fiction'' Wired pp. 9-10 (2023-08) 31-08</ref>

What we must focus on is the UX and its effect on the people that interact with it.

===The Smart Agent===
Now that an A.I. can apply reason to problems, it should be expected that such agents on on the horizon. That prediction was made in 2021, and there seems to be glimmerings of AI appearing in the user agents in our hands in 2023, but it is still all driven by Large Language Models (LLM) in the cloud. Intel is predicting that user devices in 2024 will include and AI. Loading LLM into [[Edge Computing]] devices are needed before AI will work in small devices.

Nvidia has been building GPUs for personal computers for a long time before they were purposed for [[Artificial Intelligence]]. But on 2023-12-14 they were not used for that on personal computers when Intel unveiled [https://www.intel.com/content/www/us/en/newsroom/news/ai-everywhere-core-ultra-5th-gen-xeon-news.html#gs.26279f "Core Ultra and 5th Gen Xeon – that will change how customers and consumers enable AI solutions in the data center, cloud, network, at the edge, and on the PC"] to bring AI to any home computer user with a budget for a new computer.

===Neural Nets===

Still the most popular, they appear to have reached a limit in what they can explain about human intelligence. Perhaps we need to add compassion and [[Common Sense]] back in.

The theory of neural nets has been documented in Quanta <ref>Kevin Hartnett, ''Foundations Built for a General Theory of Neural Networks.'' (2019-01-31) Quanta https://www.quantamagazine.org/foundations-built-for-a-general-theory-of-neural-networks-20190131/</ref> A description of the Turing award to the winners for applying neural nets to solving problem (as opposed to explaining intelligence) was summarized by Neil Savage<ref>Neil Savage, ''Neural Net Worth.'' '''CACM Vol 62''' no 6 p. 12 (2019-06)</ref> who specifically talks about the need for more involvement of humanities in long term solutions.

===Deep Neural Network Accelerators===
Now we come to the logical and nearly the final phase of [[Artificial Intelligence]], the design of one AI by another.<ref>Adam Zewe, ''Accelerating AI tasks while preserving data security'' MIT News (2023-10-30) https://news.mit.edu/2023/accelerating-ai-tasks-while-preserving-data-security-1030</ref> This is probably just the first step to the complete control of the AI [[Ecosystem]] by the AI themselves when the designs become to complex for human understanding.<blockquote>With the proliferation of computationally intensive machine-learning applications, such as chatbots that perform real-time language translation, device manufacturers often incorporate specialized hardware components to rapidly move and process the massive amounts of data these systems demand. Choosing the best design for these components, known as deep neural network accelerators, is challenging because they can have an enormous range of design options. This difficult problem becomes even thornier when a designer seeks to add cryptographic operations to keep data safe from attackers. Now, MIT researchers have developed a search engine that can efficiently identify optimal designs for deep neural network accelerators, that preserve data security while boosting performance.</blockquote>

===Libratus===

Libratus did not use expert domain knowledge, or human data that are specific to [the game domain]. Rather, the AI was able to analyze the game's rules and devise its own strategy. The technology thus could be applied to any number of imperfect-information games. Such hidden information is ubiquitous in real-world strategic interactions, including business negotiation, cybersecurity, finance, strategic pricing and military applications.<ref>Byron Spice, ''CSD's Sandholm, Brown To Receive Minsky Medal.'' (2018-11-06) Carnegie-Mellon News https://www.cmu.edu/news/stories/archives/2018/november/minsky-medal.html</ref>

===Alpha Zero===

After Alpha Go, the first version to use [[Matrix Calculation|Tensor Processing Units]] (TPU) defeating the best human GO play in 2016,<ref>McMillan, Robert. "Google Isn't Playing Games With New Chip". (2016-05-18) The Wall Street Journal. Archived from the original on 29 June 2016. </ref> the Google Deep Mind project Alpha Go was generalized into a far simpler version Alpha Zero<ref>Steven Strogatz, ''One Giant Step for a Chess-Playing Machine.'' (2018-12-26) New York Times https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html</ref> that was just given the rules of the game, whether Go or Chess, and allowed to play against another instance of itself until it had mastered the subject to be able to beat any existing computer model. Garry Kasparov, the world chess said that Alpha Go developed a style of play that "reflects the truth" of the game rather than "the priorities and prejudices of programmers." That pretty much ends the discussion about the superiority of "Intelligent Design" which turns out to be just another myth that humans tell about themselves and their anthropomorphic gods. It also ended the testing of [[Artificial Intelligence]] against humans in board games. There was very little generally newsworthy announcements about [[Artificial Intelligence]] for the next 6 years.

===Open AI===
Open AI Is a research institute founded by Elon Musk and Sam Altman to make AI results available to everyone. Their finding (which echoed that of Deep Mind at Google) in early 2019 was that this is a very bad idea. Like any technology it can ready be turned to evil purposes as reported in Wired.<ref>Alyssa Foote, ''The AI Text Generator That's Too Dangerous to Make Public.'' (219-02-19) Wired Magazine https://www.wired.com/story/ai-text-generator-too-dangerous-to-make-public/</ref> This finding should certain give pause to any company that seeks to use AI for their own purposes. It is too easy for AI to be turned to the purposed of evil intent. Sam Altman wrote an essay ''Moore's Law for Everything''<ref>Sam Altman ''Moore's Law for Everything'' personal blog (2021) https://moores.samaltman.com/</ref> that was reviewed by Ezra Klein.<ref>Ezra Klein ''Sam Altman on the A.I. Revolution, Trillionaires and the Future of Political Power'' NY Time (2021-06-11) https://www.nytimes.com/2021/06/11/opinion/ezra-klein-podcast-sam-altman.html?action=click&module=Opinion&pgtype=Homepage</ref>

Not everyone is captivated by Open AI. Quote from the New Yorker: [https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web "OpenAI’s chatbot offers paraphrases, whereas Google offers quotes. Which do we prefer?"] To be fair, the version called Copilot by Microsoft does provide links to the original documents.

==Ethics==
Most of the effort in protecting the human population from the dangers of [[Artificial Intelligence]] focus on arguments about ethics. [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Quandaries in AI-ML is one of a growing list on YouTube] joined by an ACM webinar on [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Challenges in AI]. What remains unclear is how AI ethics can be separated from [[Censorship]]. Most of the effort at training AI involves a list of taboo subjects. The problem here is that not everyone agrees on what ethics should apply. For example, is abortion in one country to be considered to be murder and in another country just an exercise in free will. It is unclear whether a discussion of abortion methods, suicide or similar health issues is an ethical or a moral discussion. Similar issues are easy to list, such as free speech or social credit. It comes down to a question of whether ethics in AI will ever be any more that just censorship that is inconsistently applied?

===In Healthcare===
* [https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Forange.hosting.lsoft.com%2Ftrk%2Fclick%3Fref%3Dznwrbbrs9_6-2be2bx32c520x081583%26&data=04%7C01%7C%7C1b0cfc8a7db9484bb50308d946e0a1e9%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637618752869025374%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PaRyaYQadjyHbH5wP%2FruYlnD0B5rftISxc8a9dfAJeQ%3D&reserved=0 WHO Releases AI Guidelines for Health Government Technology] 2021-07-12 <blockquote>A new report from the World Health Organization (WHO) offers guidance for the ethical use of artificial intelligence (AI) in the health sector. The six primary principles for the use of AI as set forth in the report are to protect autonomy; promote human well-being, safety, and the public interest; ensure transparency, explainability, and intelligibility; foster [[Responsibility]] and accountability; ensure inclusiveness and equity; and promote responsive and sustainable AI. These principles are intended as a foundation for AI stakeholders, including governments, developers, and society. The report, Ethics and Governance of Artificial Intelligence for Health, finds public- and private-sector collaboration necessary to ensure AI’s accountability.<blockquote>

===Back to the Future===
Brian Christian described "The Alignment Problem"<ref>Brian Christian ''The Alignment Problem'' W. W. Norton (2020-10-06) ISBN 978-0-393-86833-3</ref> as the difficulty that AI has with aligning the results provided by the AI with society's goals. It is worse than you might imagine. The results from AI as of 2020 have not proven to be appropriate. This is, in part, because they are trained on data from social systems that are also deemed to be inappropriate. What most researchers are doing from this point is trying to determine how natual intelligence develops and feeding that back into the AI systems.

The first person to try to understand "The Construction of [[Reality]] in the Child' was Jean Piaget<ref>Jean Piaget ''The Construction of Reality in the Child'' (original 1954) Ballentine ISBN 345-02338-2-165</ref>. The first line of the book is "To understand how the budding intelligence constructs the external world, we must ask whether the child, in its first months of life, conceives and perceives things as we do, as object that have substance, that are permanent and of constant dimensions." Now that NMR brain scanners can image a thought within an infant's mind<ref>Rachel Fritts, ''How do Babies perceive the World?'' MIT News July/August 2021 p 24 ff</ref>, we can actually imagine that the brain really is just the Chinese Room that so perplexed Searle.<ref>James Somers, ''Head Space'' The New Yorker Dec. 6, 2021, p 30 ff.</ref> Searle's objection to [[Artificial Intelligence]] was that it was just a symbol manipulation scheme and that cannot be all that is involved in understanding. Interestingly at the same time Noam Chomsky created a generative linguistic model of language that was nothing more than the manipulation of symbols.<ref>George Lakof, ''The Functionalist's Dilemma'', American Scientist https://www.americanscientist.org/article/the-functionalists-dilemma</ref> As you might imagine, they disagreed often in print.<ref>''Chomsky’s Revolution: An Exchange'' (2002-07-18) https://www.nybooks.com/articles/2002/07/18/chomskys-revolution-an-exchange/</ref> See the wiki page on [[Knowledge]] for more on this dispute. Andy Clark brings together ideas and techniques from robotics, neuroscience, infant psychology, and artificial intelligence. He addresses a broad range of adaptive behaviors, from cockroach locomotion to the role of linguistic artifacts in higher-level thought.<ref>Andy Clark, ''Being There: Putting Brain, Body, and World Together Again'' 1996 https://direct.mit.edu/books/book/3917/Being-TherePutting-Brain-Body-and-World-Together</ref>

And finally, the AI research community is recognizing the turn of AI towards Natural Science and is not too happy about that?<ref>Subbarao Kambhampati, ''AI as (an Ersatz) Natural Science.'' '''CACM 65''' No 9 (2022-09) pp 8-9</ref> Kambhampati reports that "Indeed, in a 2003 lecture, Turing laureate Lesle Lamport sounded alarms about the very possibility of the future of computing belong to biology rather than logic, saying it will lead us to living in a world of homeopathy and faith healing<ref>Leslie Lamport, ''The Future of Computing: Logic or Biology'' (2003-07-23) https://bit.ly/3ahrsaI</ref> Perhaps we should go back to one of the authors of the ultimate book on logic, the Principia Mathematica Alfred North Whitehead<ref>Alfred North Whitehead, ''Principia Mathematica'' Cambridge UP (Second edition 1927)</ref> who said "Metaphysical categories are not dogmatic statements of the obvious, they are tentative formulation of the ultimate generalities. If we consider any scheme of philosophic categories as one complex assertion, and apply it to the logician's alternative, true or false, the answer must be that the scheme is false. The same answer must be given to like question respecting the existing formulate principles of any science."<ref>Alfred North Whitehead, ''Process and Reality'' Free Press (1929)</ref>

==Truth from the AI==
<pre>
Lying requires belief + intention.
* Deception requires a goal to induce false belief.
* AI has neither. So by definition, it can’t be “lying” or “deceiving” - at least not in the human sense.

Yet the desire to interpret internal model states as belief-like is so strong, researchers are now building probes to "extract" them.
Words like “belief,” “reasoning,” “understanding,” "thought", and “intelligence” are being borrowed from human cognition - without being earned, and devaluing them all in the process. Not just by journalists, or AI researchers, but especially the ones hyping the tech.

Why?
* Not just to replace workers and cut wages - they know if we have no money, we can’t consume. The system collapses.
* This is about reshaping access to knowledge itself. Controlling what you see. How you learn. Who you become. Who you obey. Who you trust.

</pre>

===Effective Altruism===
Compare this with wiki page on [[Expected Value]].

Under the concept that those with wealth should be careful that the resources that the contribute back to society have a positive impact. In other words, they want to control the way their money is used. An effort to impact AI in this way is underway. <ref>Timnit Gerbu ''Effective Altruism Is Pushing a Dangerous Brand of "AI Safety"'' Wired (2022-11-30) https://www.wired.com/story/effective-altruism-artificial-intelligence-sam-bankman-fried/</ref><blockquote>EA is defined by the Center for Effective Altruism as “an intellectual project, using evidence and reason to figure out how to benefit others as much as possible.” And “evidence and reason” have led many EAs to conclude that the most pressing problem in the world is preventing an apocalypse where an artificially generally intelligent being (AGI) created by humans exterminates us. To prevent this apocalypse, Some of the billionaires who have committed significant funds to this goal include Elon Musk, Vitalik Buterin, Ben Delo, Jaan Tallinn, Peter Thiel, Dustin Muskovitz, and Sam Bankman-Fried, who was one of EA’s largest funders until the recent bankruptcy of his FTX cryptocurrency platform. As a result, all of this money has shaped the field of AI and its priorities in ways that harm people in marginalized groups while purporting to work on “beneficial artificial general intelligence” that will bring techno utopia for humanity. This is yet another example of how our technological future is not a linear march toward progress but one that is determined by those who have the money and influence to control it. </blockquote>

Nietzshe had a very dismal view on the efficacy of altruism<ref>Friedrich Nietzsche. ''The Twilight of the Idols'' no 35</ref> <blockquote>An "altruistic" morality, a morality under which selfishness withers, is in all circumstances a bad sign. This is true of individuals and above all of nations. The best are lacking when selfishness begins to be lacking. Instinctively to select that which is harmful to one, to be lured by "disinterested' motives,—these things almost provide the formula for decadence. "Not to have one's own interests at heart" —this is simply a moral fig-leaf concealing a very different fact, a physiological one, to no longer know how to find what is to my interest."... Disintegration of the instincts!—AII is up with man when he becomes altruistic.— instead of saying ingenuously "I am no longer any good," the lie Of morality in the decadent's mouth says: "Nothing is any good,—life is no good."—A judgment of this kind ultimately becomes a great danger; for it is infectious, and it soon flourishes on the soil of society...</blockquote>

A better answer to AI might be to allow those who work directly with underserved people. The best way to measure the success of a society is to see how it treats the least of its members.<ref>Karen Hao ''A new vision of artificial intelligence for the people'' Technology Review (2022-04-22)</ref><blockquote>This story is the fourth and final part of MIT Technology Review’s series on AI colonialism, the idea that artificial intelligence is creating a new colonial world order. The project is a radical departure from the way the AI industry typically operates. Over the last decade, AI researchers have pushed the field to new limits with the dogma “More is more”: Amass more data to produce bigger models (algorithms trained on said data) to produce better results. The approach has led to remarkable breakthroughs—but to costs as well. Companies have relentlessly mined people for their faces, voices, and behaviors to enrich bottom lines. And models built by averaging data from entire populations have sidelined minority and marginalized communities even as they are disproportionately subjected to the technology’s impacts. Over the years, a growing chorus of experts have argued that these impacts are '''repeating the patterns of colonial history.''' Global AI development, they say, is impoverishing communities and countries that don’t have a say in its development—the same communities and countries already impoverished by former colonial empires.</blockquote>

See the wiki [[Social Justice]] for more details include Herbert Simon on Bounded Rationality.

==Regulation==
The big question is it safe? The simple answer is that new technology always comes with new risks. Antitrust law is the best-known method for controlling enterprises too large to be controlled by the government. Perhaps they can be used to control a super intelligent being of another sort. Perhaps that's all an AI ever is, corporations with computers and imposed limitations.
* 2024-10-24 [https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/ Memorandum on Advancing the United States’ Leadership in Artificial Intelligence; Harnessing Artificial Intelligence to Fulfill National Security Objectives; and Fostering the Safety, Security, and Trustworthiness of Artificial Intelligence]
* NIST issued a call for participation on [https://www.federalregister.gov/documents/2023/11/02/2023-24216/artificial-intelligence-safety-institute-consortium Artificial Intelligence Safety Institute Consortium] 2023-11-02 in the federal resister.
* President Bidin issued a [https://www.whitehouse.gov/ostp/ai-bill-of-rights/ Blueprint for an AI Bill of Rights] (2022-10-04) problems are well documented. In America and around the world, systems supposed to help with patient care have proven unsafe, ineffective, or biased. Algorithms used in hiring and credit decisions have been found to reflect and reproduce existing unwanted inequities or embed new harmful bias and discrimination. Unchecked social media data collection has been used to threaten people’s opportunities, undermine their privacy, or pervasively track their activity—often without their knowledge or consent. Five Principles:
# Safe and Effective Systems - You should be protected from unsafe or ineffective systems.
# Algorithmic Discrimination Protections - You should not face discrimination by algorithms and systems should be used and designed in an equitable way.
# Data Privacy - You should be protected from abusive data practices via built-in protections and you should have [[Agency]] over how data about you is used.
# Notice and Explanation- You should know that an automated system is being used and understand how and why it contributes to outcomes that impact you.
# Human Alternatives, Consideration, and Fallback - You should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter.
* A law that went into effect in New York City in January 2023<ref>Richard Vanderford, ''New York AI Bias Law Prompts Uncertainty'' Wall Street Journal (2022-09-21) </ref> requires companies to audit their artificial intelligence (AI) hiring systems to assess whether they incorporate racial and gender biases. The law holds hiring companies liable for any biases and could impose fines for violations, but lacks clear guidelines for the AI audit process. While the city’s Department of Consumer and Worker Protection has not offered a timeline for when it will publish rules to implement the law, some companies already are taking steps to comply. Said Anthony Habayeb of AI governance software company Monitaur Inc., "Instead of waiting for someone to tell me what to do… I built controls around these applications because I know like with any software, things can and do go wrong."

==References==

<references />

===Other Material===
* See the wiki page on [[Intelligent Goals]].
* See the wiki page on [[Machine Learning]].
* See the wiki page on [[Fairness Accountability Transparency Ethics]]
* See the wiki page on [[Genetic Programming]]

[[Category: Glossary]]
[[Category: Model]]
[[Category: Authentication]]
[[Category: Artificial Intelligence]]

Artificial Intelligence

2026-03-23T00:37:08Z

Tom: /* Creativity */

==Full Title or Meme==
Intelligence without human lived experience or human compassion.

"The future is already here — it's just not very evenly distributed" - William Gibson.

The term [[Artificial Intelligence]] was coined, in part, out of spite. In 1955, the Dartmouth professor John McCarthy assembled a proposal for what became the first-ever conference on AI. McCarthy refused to use the prevailing term, "cybernetics": That would have necessitated inviting Norbert Wiener, then the world's leading cyberneticist, whom he found insufferable. Instead, McCarthy minted a new term that took off in the digital era, taking him along as a founder of the field. <ref>''Fun fact: Cybernetics'' Technology Review (2025-08) p 11</ref>

==Context==
===The Broad concepts===
* There are two great forces both in physics and in human history
** The harnessing of energy which started with the use of animal power to human benefit.
** The harnessing of information which started with philosophy and then Gutenberg's press.
* Likewise there are two great revolutions in human society.
** The Industrial revolution focused on control and use of energy which peaked in the 20th century.
** The Information revolution focused on control and use of information which is peaking in the 21st century.
* The social changes to accompany these revolution had to before the revolutions could complete.
** The educational system needed to be extended to all who work to engineer the control of energy.
** Now we need a second shift to educate most of the workers to control the information needed.
* The use of energy and information in harmony with the reality of the Earth's resources is a challenge.
** The profligate use of energy has altered the environment in ways that threaten human existence.
** The profligate use of information may prove to alter the environment of work to be a bigger threat.
* The struggle between conservative and progress forces in politics has been with us for ever.
** The conservative mind wants to hold onto the privilege of the past and present that they enjoy.
** The progressive mind want so shape the future to make that future a better place for us all.
* The Collingridge dilemma describes the challenge of regulating new technologies.
** Early stage: When a technology is new, its future impact is unclear, making regulation difficult.
** Later stage: Once the technology is widely adopted, it becomes harder to control or modify.

===Why Bother?===
Some jobs are too risky.
Some tasks are too mind-numbing.
Some conditions are too toxic.
Some decisions are too complex.

AI should be our answer to protect, preserve, and uplift human potential, not disregard or displace it.

===What is Intelligence===
* Here Intelligence relates primarily to [[Identity]] [[Knowledge]], but that has many general aspects.
* In 1950 Turing asked "Can Machines Think".<ref>A. M. Turing, ''COMPUTING MACHINERY AND INTELLIGENCE'' 1950-10-01 https://academic.oup.com/mind/article/LIX/236/433/986238</ref>
* People have been working on artificial intelligence starting with a conference at Dartmouth College in the 1950s. There, computer scientists and mathematicians set a very ambitious goal: to develop computers that could understand language, translate it, see patterns, and more.
* Marvin Minsky was teaching undergrad courses in 1964 using the book "Computers and Thought."<ref>Edward Feigenbaum, & Julian Feldman, joint ed ''Computers and Thought'' https://archive.org/details/computersthought00feig</ref>
* in 2023 Bill Gates in "AI is about to completely change how you use computers" said<ref>Bill Gates ''AI is about to completely change how you use computers'' https://www.gatesnotes.com/AI-agents</ref> "I still love software as much today as I did when Paul Allen and I started Microsoft. But—even though it has improved a lot in the decades since then—in many ways, software is still pretty dumb.:
* Some people think that [[Artificial Intelligence]] must needs be cold or sterile, but we have plenty of [[Evidence]] that is not so.
** Elisa was the first but [https://arstechnica.com/information-technology/2023/12/real-humans-appeared-human-63-of-the-time-in-recent-turing-test-ai-study/ it best chat gpt so I guess the turning test isn't so useful]
** Alexa works on Amazon echo devices and is constantly updated so there is nothing stable to compare.
** GPT was getting some press with GPT-2 included in Technology Review tagged it as a break-through technology in 2021<ref>Will Douglas, '' GPT-3'' Technology Review '''Vol 124''' No 2 (2021-03) pp. 34-35</ref>, but nothing managed to get public [[Attention]] until... {See below for more on Open AI}
** Chat-GPT which was the first publicly accessible product that was not completely cringe-worthy on 2022-11-30.
* Is [[Artificial Intelligence]] even [https://www.linkedin.com/pulse/machine-intelligence-artificial-sean-manion-phd-oz91e/ artificial]? Or is it just yet another way to create a balance between [[Chaos and Order]].
* Determining what [[Artificial Intelligence]] actually is and whether is is like human intelligence turns out to be a impossible task. Ray Kurzweil said<ref>Ray Kurzweil, ''The Singularity is Nearer''</ref><blockquote>The reason a well-defined empirical test is necessary is that ... humans have a powerful tendency to redefine whatever artificial intelligence achieves is not ''really'' so hard in hindsight. This is often referred to as the "AI effect". ... Intuitively [the hallucinations and odd answers] feels like a problem. It's tempting to think that Watson "ought" to reason like humans do. But i would propose that this is superstition. In the real world, what matters is how an intelligent being acts.</blockquote>
===Origin===
Over the summer of 1956 a small but illustrious group gathered at Dartmouth College in New Hampshire; it included Claude Shannon, the begetter of information theory, and Herbert A. Simon, the only person ever to win both the Nobel Memorial Prize in Economic Sciences awarded by the Royal Swedish Academy of Sciences and the Turing Award awarded by the Association for Computing Machinery. They had been called together by a young researcher, John McCarthy, who wanted to discuss “how to make machines use language, form abstractions and concepts” and “solve kinds of problems now reserved for humans”. It was the first academic gathering devoted to what McCarthy dubbed “artificial intelligence”. And it set a template for the field’s next 60-odd years in coming up with no advances on a par with its ambitions.<ref>John McCarthy, M. Minsky, N. Rochester, C.E Shannon, ''A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence'' (1955-08) http://raysolomonoff.com/dartmouth/boxa/dart564props.pdf</ref>

During 1955 and 1956 computer scientist and cognitive psychologist Allen Newell, political scientist, economist and sociologist Herbert A. Simon, and systems programmer John Clifford Shaw, all working at the Rand Corporation in Santa Monica, California, developed the Logic Theorist, the first program deliberately engineered to mimic the problem solving skills of a human being. They decided to write a program that could prove theorems in the propositional calculus like those in Principia Mathematica by Alfred North Whitehead and Bertrand Russell.<ref>''Newell, Simon & Shaw Develop the First Artificial Intelligence Program'' https://www.historyofinformation.com/detail.php?id=742</ref>

In 1956 John von Neumann agreed to present a discussion on "The Computer and the Brain" as the Yale Silliman lecture. He was too ill to complete the full gamut of lectures but did get enough presented to be published by Yale.<ref>John von Neumann, ''The Computer and the Brain'' (1958) Yale UP ISBN 0300024150</ref> Perhaps the most interesting part is the discussion of the difference between arithmetic and logic. He noted that only a few bits of information were available to make a choice between options, but that the computer's arithmetic was growing to every more bits, now 64 in most machines, to give sufficient precision to make a choice. Only machines defined after 2020 used ultra-low precision (2–4 bits) primarily for inference, not training. IBM researchers presented techniques for 4-bit quantization at NeurIPS 2020, focusing on reducing memory and compute costs while maintaining accuracy for deep neural networks

The first international conference was "Mechanisation of Thought Processes"<ref>National Physical Laboratory at Teddington,''Mechanisation of Thought Processes'' https://aitopics.org/download/aiclassics:98875BCD</ref> was held in 1958 with presentations by Minsky and John McCarthy.

===Perceptron===
Also called a McCulloch-Pitts [[Neuron]] is an attempt to mimic an animal [[Neuron]], originally with a purpose built computer system. While it was first described in 1947, it was not realized until Frank Rosenblatt built is Mark I computer using perceptrons as a primary physical element. He continued to build larger machines with more layer until his early death in 1971. Minskey and Papert wrote an entire book<ref> M. Minsky & S. Papert, ''Perceptrons'' (1987-12-28) ISBN 978-0262631112</ref> where they purported to show that it was a dead-end idea which basically blocked progress on them til the 1980's. The Large Language Models that became popular in the early 2020's are based on Multi-Level Perceptrons (MLP).

===End Game===
Now i am become death, the destroyer of worlds. Quoted by Robert Oppenheimer at the completion of the Atomic bomb and in the movie "Ex Machina" from the Bhagavad Gita.

Perhaps a good question to ask is, "Can Intelligence Be Separated from the Body?"<ref>Oliver Whang, ''Can Intelligence Be Separated From the Body?'' New York Times (2023-04-11) https://www.nytimes.com/2023/04/11/science/artificial-intelligence-body-robots.html?</ref><blockquote>Maybe the mind is like a video game controller, moving the body around the world, taking it on joy rides. Or maybe the body manipulates the mind with hunger, sleepiness and anxiety, something like a river steering a canoe. Is the mind like electromagnetic waves, flickering in and out of our light-bulb bodies? Or is the mind a car on the road? A ghost in the machine?</blockquote>

Perhaps a better question to ask is, "Can Intelligence Be Separated from a persistent Identifier"?<blockquote>The [[Identifier]]s used with AI in 2023 are basically just service [[Identifier]]s. in 2023 at the beginning of the general availability of the deployments of AIs that did not embarrass their creators, was also the time that context was provided to the AI to help it understand a thread of a conversation. Before that every query was de novo, without context. But still there is not a persistent history of conversations, or any concept that that is an "I" there to establish a history, which is still the basis of trust. Perhaps we must wait until we are able to crate an AI that persists an [[Identity]], we, as individuals, will be able to establish trust with it, as an individual. </blockquote>

==Problems==
* Humans seem to not be able to use [[Intelligent Design]] alone to fashion [[Artificial Intelligence]]. In fact researchers continue to go back to nature and human functioning for inspiration and even for algorithms. In other words, our understanding of the meaning of the term [[Artificial Intelligence]] has been and will continue to be evolving over time.
* Nature itself since the Big Bang seems to be based on [[Self-organization]] for its framework.
* Training an [[Artificial Intelligence]] with human behaviors will result in unacceptable behavior by the [[Artificial Intelligence]]. The most common complaint is real, or imagined, or invented, bias against one group or another.
** Microsoft released Tay,<ref>Peter Lee, Learning from Tay’s introduction. (2016-05-25) https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/</ref> a web robot (bot) to respond to tweets and chats<ref>Sarah Parez, ''Microsoft silences its new A.I. bot Tay, after Twitter users teach it racism.'' (2018-03-24) Tech Crunch https://techcrunch.com/2016/03/24/microsoft-silences-its-new-a-i-bot-tay-after-twitter-users-teach-it-racism/</ref> in May 2016. The result was disastrous as Tay learned to be racist from its all-too-human trainers and was shut down in days.
** Google has been plagued with reports and legal action on its search results nearly continuously since it was introduced; in 2019 the president, Donald Trump, accusing it of favoritism to leftist causes.<ref>Farhad Manjoo, ''Search Bias, Blind Spots And Google.'' (2018-08-31) New York Times p. B1</ref> Researcher Safiya U. Noble has written a book<ref>Safiya U. Noble, ''Algorithms of Oppression: How Search Engines Reinforce Racism.'' (2018-02-20) ISBN 978-1479849949</ref> mostly complaining that all-too-human programmers injected their own prejudices into their work. What else could he expect of humans, to rise above them selves, whatever that might mean in terms of free speech or freedom of religion?
** The page [[Right to be Forgotten]] describes an effort in Europe to teach the search engines to with-hold information about people that they don't like by collecting all of the information that they don't like. Be careful what you ask your [[Artificial Intelligence]] to do for you; it might just spill the beans some day, perhaps under court order.
** In the movie "Blade Runner 2049" the protagonist's AI girl friend asks to have her memory wiped so that the police cannot compel her to testify against him. One would hope that our future AIs will be that compassionate towards us; but that kind of compassionate behavior will probably be illegal.
===Personhood===
This is a really bad idea. It's solving a problem from the last generation of wallets when we need something that will work when the wallet is just an AI agent that can mimic anything thrown at it. Face it - the cred even today comes from a wallet that needs to prove its own creds before it can say anything about who is (or is not) at the controls. Face it - the internet of the future is just ai agents talking to each other.
* The development of full [[Artificial Intelligence]] could spell the end of the human race. - Stephan Hawking BBC Interview 2014. There are lots of dystopian futures depicted by science fiction authors. Some, like the Terminator series seem unlikely. The scenario in the movie Gattaca where humans were separated into categories based on the genome has already been enabled in the Personhood credential<ref>Pymnts, ''AI’s Human Mimicry Spurs ‘Personhood Credential’ Proposal'' 2024-09-24 https://www.pymnts.com/digital-identity/2024/ais-human-mimicry-spurs-personhood-credential-proposal/</ref> which would allow machines to determine by themselves who is qualified to be classified as human.
* Descartes claimed "I think therefore I am". But it seems that we need to be careful about applying that idea to [[Artificial Intelligence]]. Although his philosophy is no longer considered to be valid even for humans.
* [https://workshop.vennfactory.com/p/personhood-the-killer-credential Personhood: The Killer Credential?] Digging deeper into a very human challenge - Eve Maler 2024-09-16

===Vulnerabilities===
* 2024-08-20 [https://www.w3.org/reports/ai-web-impact/#safety-and-security AI & the Web: Understanding and managing the impact of Machine Learning models on the Web] from the W3C
* [https://home.treasury.gov/news/press-releases/jy2212U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector] U.S. Department of the Treasury 2024-03-27
* [https://arxiv.org/pdf/2402.00888.pdf Security and Privacy Challenges of Large Language Models: A Survey] (2024-02)
* Input to the model used by the AI come from two sources, each of which can be compromised.
# Training data from the Internet can be tampered with to give bad results.
## For starters, the internet data is typically censored to remove input that would cause embarrassment and [[Conduct Risk]] for the sponsors.
## Some locales will censor data to make themselves look good or promote some social point of view of the government of of the industry.
# The input from a single user can cause unexpected changes to later results.
## The early AI release were compromised with a variety of hate speech that quickly resulted in really nasty output from the AI.
## Artists seeking to protect their intellectual property from later output have learned to put small changes in the images that cause them to be mislabeled or worse.<ref>Melissa Hakkiliia, ''This new data-poisoning tool lets artists fight back against generative AI'' MIT Technology Review 127 No 1 (2024-01) p 7=8</ref>
* Giving a AI a goal may result in unexpected and undesired solutions.
** Bruce Schneier described<ref>Bruce Schneier, ''Hackers Used to Be Humans. Soon, AIs Will Hack Humanity'' Wired (2021-04-18) https://www.wired.com/story/opinion-hackers-used-to-be-humans-soon-ais-will-hack-humanity/?bxid=5c5b250d24c17c67f8640083</ref> AI to be Like crafty genies. They will grant our wishes and then hack them, exploiting our social, political, and economic systems like never before.

See the wiki on [[AI Threat Model]]s for more details on this subject.

===Uniquely Human===
Neither the Turning Test, nor any other known test, is designed to determine if an [[Artificial Intelligence]] is really human. The best way to understand what natural intelligence is would be to look at the reason human evolved to have such a large brain. There appear to be these principle requirements:<ref>Edward O. Wilson, ''The Meaning of Human Existences'' Liveright (2014) Chapter 2 ISBN 9780871401007</ref>
# Necessity for a large brain is the changing [[Ecosystem]] that a small remnant of Homo Sapiens living in the south of Africa needed in order to survive
# Changing locations to track new sources of protein and the switch from plant to animal protein.
# Empathy to be able to understand what is in someone else's mind to deduce if they notice you and if they think you are targeted by them as prey.
The results of these were two processes that served to enable survival in such varying [[Ecosystem]]s/
# Passing learning from one generation to the next.
# An extended adolescence with mothers that ceased being fecund to tend to their existing children and the extended life of grandparents to help with stories and other techniques to bring the knowledge of the tribe to the growing child.
The search for causes that were more complex than the ability of genes to carry information, such as the migratory pattern of birds is described with more detail in the wiki page on [[Chance and Necessity]].

* [https://www.linkedin.com/pulse/machine-intelligence-artificial-part-4-cybernetics-sean-manion-phd-kfi0e Machine Intelligence is not Artificial - Part 4: Cybernetics and Norbert Wiener]<blockquote>What is of particular note is not simply the existence of cybernetics as a discipline dating back to the 1940s, but the broad popularity Wiener's book and the topic received not only from scientists and engineers, but also the general public. His 1948 book went through four printings in the first 6 months and was a best seller, rivaling the contemporary Kinsey Report in popularity over the following decade. This was despite this book, ostensibly for a lay audience, being extremely dense with many chapters full of mathematical equations. One of the chapters in the original version, "The Computing Machines and the Nervous System," makes it clear that Wiener's mention of humans AND machines in the book's subtitle was pointing towards the comparison of these systems, making this book one of the early popularizers of the idea of "thinking machines" (but by no means the only one). This was two years before Alan Turing's 1950 paper "Computing Machinery and Intelligence," which gave us the Turing test for AI.</blockquote>

* Human-centered AI - [https://aiindex.stanford.edu/report/?utm_source=Stanford+HAI&utm_campaign=525788e5b7-AI_INDEX_2024_EMAIL_CAMPAIGN&utm_medium=email&utm_term=0_-525788e5b7-%5BLIST_EMAIL_ID%5D&mc_cid=525788e5b7&mc_eid=e55581d782 THE AI INDEX REPORT - Measuring trends in AI] 2024 HAI Stanford

===Explaining AI===
Most users want to know '''WHY''' a decision was make or an action was performed. The article "Solving for Why"<ref> Marina Kraskovsky, ''Solving for Why'' '''CACM 65''' No 2 (2202-02) pp 11ff</ref> is a focus on [[Causality]] to help researchers overcome shortcomings that have bedeviled more traditional approaches to AI. It is not enough to predict things, a good AI will need to solves things and explain why the choice was make if the decision is to have any support by the human population. People will ask if the decision was fair and will second guess the choices make by the AI unless they understand and accept the decision. Today AI decision are fragile, a slight change in the environment may have drastic changes in outcomes. This may well be inexplicable to humans. It will be insufficient to convince experts in the field about the choices made. It is apparent in 2022 that expert opinions about COVID have not been very popular. Even more evidence will be required of a AI decision. This will be particularly true when the decision is costly in any way. Humans like to have choices, which may require the AI to offer choices. That will have the benefit of giving the humans some feeling that they are in control. But beware of "Hobson's choice". That was a stable master at Oxford that was required to give the students a choice of horses, but selected the options so that there was no real choice.

===Creativity===
the idea that early large language models (LLMs) showed more creative writing, and that this quality degraded as rules, alignment, and post‑training constraints were added. I’ll separate what is documented, what is strongly suggested by reporting, and what remains a hypothesis, and I’ll anchor claims to explicit sources; especially the recent work of Jasmine Sun reports a striking consensus among AI researchers and writers she interviewed:
* Earlier models (especially GPT‑2–era systems) often produced unexpected, playful, and strange continuations.
* Modern models are more fluent and correct—but less surprising and less interesting to read.

Sun explicitly reports examples from researchers and poets who experimented with GPT‑2, describing outputs that wandered, took risks, and made odd but evocative associations that newer models no longer produce. This is not nostalgia alone: Sun notes that AI executives themselves admit they have not released a model that writes well, despite enormous gains in capability.

Jasmine Sun is currently The Atlantic’s clearest voice arguing that AI degrades creativity by eliminating stakes, judgment, and responsibility, even as it perfects surface‑level fluency.<ref>Sasmine Sun, ''The Human Skill That Eludes AI'' The Atlantic 2026-03-17</ref>

===Paradox===
Moravec’s Paradox describes our experience that it is hard to teach machines to do things that are easy for most humans (walking, for example) but comparatively easy to teach them things that are challenging for most humans (chess comes to mind).

Polanyi’s Paradox is named for the philosopher Michael Polanyi (1891-1976), who developed the concept of “tacit knowledge”: Central to Michael Polanyi’s thinking was the belief that creative acts (especially acts of discovery) are shot-through or charged with strong personal feelings and commitments (hence the title of his most famous work Personal Knowledge). Arguing against the then dominant position that science was somehow value-free, Michael Polanyi sought to bring into creative tension a concern with reasoned and critical interrogation with other, more ‘tacit’, forms of knowing. Polanyi’s argument was that the informed guesses, hunches and imaginings that are part of exploratory acts are motivated by what he describes as ‘passions’. They might well be aimed at discovering ‘truth’, but they are not necessarily [[Formal Model]]s that can be stated in propositional or formal terms. As Michael Polanyi (1967: 4) wrote in The Tacit Dimension, we should start from the fact that ‘we can know more than we can tell‘.<ref>MARK K. SMITH, “MICHAEL POLANYI AND TACIT KNOWLEDGE” AT INFED</ref>

* For synonyms to tacit knowledge see the wiki page on [[Common Sense]].
* It seems, in 2023, that generative AI is able to overcome the lack of imagination with simple random guessing where knowledge is lacking.

===Fairness===
* What is called "Algorithmic Bias" today is just the result of training any [[Artificial Intelligence]] using any data based on human behavior which typically is biased.
* We now expect more ''Fairness'' from our [[Artificial Intelligence]] as we expect from normal humans.
* "The burgeoning field of algorithmic fairness, part of the much broader field of responsible computing, is aiming to remedy the situation"<ref> Marina Krakovsky, '' Formalizing Fairness'' '''CACM 65''' No. 8 (2022-08) p 11ff.</ref> which is the result of using available data.
* Scientists have found that the English word "fairness" has many meanings, and do they would prefer terms like technical disparities, that they can define precisely. The problem with that is that what society demands is "fairness". Which means that scientists could never deliver what society demands.
* As was demonstrated in the COMPASS program to predict recidivism in convicts, it is not possible create a single solution that satisfies more than one metric for "fairness".
* Explainable Fairness means can you explain it to a judge so that they can make a determination. -- Cathy O'Neil ORCAA
* Every provider wants you to think that they are meeting the presidents EO, like the guys pitching Zero Trust Network Access (ZTNA) - do not be confused, this meets no level of [[Least Privilege]] or any other attempt to limit access to people that have earned access. It is nothing but the network folk trying to climb on band wagon.
* [https://www.readynez.com/en/blog/understanding-fair-ai-a-simple-guide/ Understanding Fair AI: A Simple Guide!]
* See the wiki on [[Fairness Accountability Transparency Ethics]] for more details on this topic.

===Disruption===
We can now say what GPT can replace - shallow human conversation and mechanical activity like law or medicine clerks perform. A GPT in a cocktail party could be wittier than anyone else. A GPT could do better than a recent grad of most law or medical schools at citing existing law or medical practice. Is anyone surprised? AI is moving up the food chain from one mechanical task to the next. We learned from aircraft pilot errors that we had to have a checklist to save pilots from error. I guess we could call the current AI a check list of stuff that is already known.

The question about what is left for humans might again be of interest. It was during the industrial revolution when machines replaced human labor. The best that the pundits of the time could suggest it that humans could have more leisure time. Now machines are moving to more tasks. But somehow employment is full in 2022. There is likely to be a reconning some day when machines and their owners will disrupt the system and cause unemployment. The rich and the poor will, yet again, come into conflict and a revolution of some sort will be required.

===Bias===
* Clearly any [[Artificial Intelligence]] is biased by its input data. Social pressures have forced any [[Entity]] that offers an AI to make if conform to the laws and regulations of the state. All countries have proscribed some set of topics or responses as unacceptable. In other words all AI are censored by the training data or the laws of the land. Even the language used for training can create biases.<ref>Antony Chayka and Andrei Sukhov, ''Comparing Chatbots Trained in Different Languages'' '''CACM 66''' no 12 pp, 6-7 (2023-12)</ref><blockquote>In our opinion, the government's position is clearly taken into account in the responses of AI systems in the national language, especially when the creation of AI was funded in the tested country.</blockquote>
* We face the same problems in raising our children. If they are raised in an environment that exhibits bias, it will be hard to change their behavior later in life.<ref>Will Douglas Heaven, ''Six Big Questions for Generative AI: Hope, Hype, Nope'' Technology Review (2024-01) p. 30ff. </ref> But if we use synthetic data to train the AI there will be huge objections as it is not possible for any training material (for either humans or AI) to be objectionable to some subset of humanity.

==Solutions==
The wiki page on [[Trust]] describes how people can get into the habit of trusting business process or technology process that really do not deserve to be trusted. The following is a listing of some of the technological solutions. The question about whether an [[Artificial Intelligence]] is trustworthy is yet to be addressed.

===Levels===
It is to be expected that level of intelligence will be applicable, which the higher levels learning from, or dependent on, simpler levels that start from basic automation and then to the following.<ref>Peter Denning and Ted Lewis, ''An AI Learning Hierarchy'' '''CACM 67''' No 12 (2024-12) p. 24ff.</ref>

# Rule-based system, these were applied to board games early in the computing age, but were not able to Handle more complex games like chess or go where experts were unable to articulate the rules by which they played.
# Supervised Learning
# Unsupervised Learning
# Generative AI
# Reinforcement Learning
# Human-machine interaction
# Aspirational AI

===Governance===
For most of 2023 the focus was on government standards for AI. The idea that trust can be initiated by government oversight is wide-spread, but wrong. Trust comes from a history of fail dealing, that is not something that governments understand or practice. But still the government can, and should, punish crimes, so the legislation should focus on how crimes committed by AI are punished. The death penalty would be one good deterrent. Here is one discussion on what governments can do:
[https://www.linkedin.com/pulse/governance-trust-guy-huntington-cus3c/ AI/bots and National Security].
===Training===
A great deal of effort is directed at the education of young children. In large part education is directed at the production of adults that a useful the the society in which they are expected to function as useful citizens. But their is a sharp disagreement on the subject. According to Ken Robinson, education should have four core purposes: personal, economic, social, and cultural <ref>Ken Robinson, Kate Robinson ''What Is Education For?'' (2022-03-02) https://www.edutopia.org/article/what-education</ref> Perhaps it is time to model the training of AI on the training of humans?

===User Experience===
The real question about the use of the current Large Language Model of [[Artificial Intelligence]] is how the device in our possession displays the interaction with the [[Artificial Intelligence]]. We humans have been reading stories and watching displays of [[Artificial Intelligence]] from the puppet Pinocchio that wanted to be a real boy to Deus ex Machina a movie about an [[Artificial Intelligence]] robot that went rouge. The common thread here is that each was a fictional account that we were able to relate to. Good authors know how to enliven the characters of a story for the reader that drags the reader into the story. The [[User Experience]] of AI for the near term will be interactions through our personal devices, even though the more interesting, and troubling, case will be when the robotic devices are not in our possession, but directly run by some gigantic corporation. The good story tellers will be the ones that grab our [[Attention]], even as is true today.<ref>Hannah H. Kim, ''A.I. is a Fiction'' Wired pp. 9-10 (2023-08) 31-08</ref>

What we must focus on is the UX and its effect on the people that interact with it.

===The Smart Agent===
Now that an A.I. can apply reason to problems, it should be expected that such agents on on the horizon. That prediction was made in 2021, and there seems to be glimmerings of AI appearing in the user agents in our hands in 2023, but it is still all driven by Large Language Models (LLM) in the cloud. Intel is predicting that user devices in 2024 will include and AI. Loading LLM into [[Edge Computing]] devices are needed before AI will work in small devices.

Nvidia has been building GPUs for personal computers for a long time before they were purposed for [[Artificial Intelligence]]. But on 2023-12-14 they were not used for that on personal computers when Intel unveiled [https://www.intel.com/content/www/us/en/newsroom/news/ai-everywhere-core-ultra-5th-gen-xeon-news.html#gs.26279f "Core Ultra and 5th Gen Xeon – that will change how customers and consumers enable AI solutions in the data center, cloud, network, at the edge, and on the PC"] to bring AI to any home computer user with a budget for a new computer.

===Neural Nets===

Still the most popular, they appear to have reached a limit in what they can explain about human intelligence. Perhaps we need to add compassion and [[Common Sense]] back in.

The theory of neural nets has been documented in Quanta <ref>Kevin Hartnett, ''Foundations Built for a General Theory of Neural Networks.'' (2019-01-31) Quanta https://www.quantamagazine.org/foundations-built-for-a-general-theory-of-neural-networks-20190131/</ref> A description of the Turing award to the winners for applying neural nets to solving problem (as opposed to explaining intelligence) was summarized by Neil Savage<ref>Neil Savage, ''Neural Net Worth.'' '''CACM Vol 62''' no 6 p. 12 (2019-06)</ref> who specifically talks about the need for more involvement of humanities in long term solutions.

===Deep Neural Network Accelerators===
Now we come to the logical and nearly the final phase of [[Artificial Intelligence]], the design of one AI by another.<ref>Adam Zewe, ''Accelerating AI tasks while preserving data security'' MIT News (2023-10-30) https://news.mit.edu/2023/accelerating-ai-tasks-while-preserving-data-security-1030</ref> This is probably just the first step to the complete control of the AI [[Ecosystem]] by the AI themselves when the designs become to complex for human understanding.<blockquote>With the proliferation of computationally intensive machine-learning applications, such as chatbots that perform real-time language translation, device manufacturers often incorporate specialized hardware components to rapidly move and process the massive amounts of data these systems demand. Choosing the best design for these components, known as deep neural network accelerators, is challenging because they can have an enormous range of design options. This difficult problem becomes even thornier when a designer seeks to add cryptographic operations to keep data safe from attackers. Now, MIT researchers have developed a search engine that can efficiently identify optimal designs for deep neural network accelerators, that preserve data security while boosting performance.</blockquote>

===Libratus===

Libratus did not use expert domain knowledge, or human data that are specific to [the game domain]. Rather, the AI was able to analyze the game's rules and devise its own strategy. The technology thus could be applied to any number of imperfect-information games. Such hidden information is ubiquitous in real-world strategic interactions, including business negotiation, cybersecurity, finance, strategic pricing and military applications.<ref>Byron Spice, ''CSD's Sandholm, Brown To Receive Minsky Medal.'' (2018-11-06) Carnegie-Mellon News https://www.cmu.edu/news/stories/archives/2018/november/minsky-medal.html</ref>

===Alpha Zero===

After Alpha Go, the first version to use [[Matrix Calculation|Tensor Processing Units]] (TPU) defeating the best human GO play in 2016,<ref>McMillan, Robert. "Google Isn't Playing Games With New Chip". (2016-05-18) The Wall Street Journal. Archived from the original on 29 June 2016. </ref> the Google Deep Mind project Alpha Go was generalized into a far simpler version Alpha Zero<ref>Steven Strogatz, ''One Giant Step for a Chess-Playing Machine.'' (2018-12-26) New York Times https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html</ref> that was just given the rules of the game, whether Go or Chess, and allowed to play against another instance of itself until it had mastered the subject to be able to beat any existing computer model. Garry Kasparov, the world chess said that Alpha Go developed a style of play that "reflects the truth" of the game rather than "the priorities and prejudices of programmers." That pretty much ends the discussion about the superiority of "Intelligent Design" which turns out to be just another myth that humans tell about themselves and their anthropomorphic gods. It also ended the testing of [[Artificial Intelligence]] against humans in board games. There was very little generally newsworthy announcements about [[Artificial Intelligence]] for the next 6 years.

===Open AI===
Open AI Is a research institute founded by Elon Musk and Sam Altman to make AI results available to everyone. Their finding (which echoed that of Deep Mind at Google) in early 2019 was that this is a very bad idea. Like any technology it can ready be turned to evil purposes as reported in Wired.<ref>Alyssa Foote, ''The AI Text Generator That's Too Dangerous to Make Public.'' (219-02-19) Wired Magazine https://www.wired.com/story/ai-text-generator-too-dangerous-to-make-public/</ref> This finding should certain give pause to any company that seeks to use AI for their own purposes. It is too easy for AI to be turned to the purposed of evil intent. Sam Altman wrote an essay ''Moore's Law for Everything''<ref>Sam Altman ''Moore's Law for Everything'' personal blog (2021) https://moores.samaltman.com/</ref> that was reviewed by Ezra Klein.<ref>Ezra Klein ''Sam Altman on the A.I. Revolution, Trillionaires and the Future of Political Power'' NY Time (2021-06-11) https://www.nytimes.com/2021/06/11/opinion/ezra-klein-podcast-sam-altman.html?action=click&module=Opinion&pgtype=Homepage</ref>

Not everyone is captivated by Open AI. Quote from the New Yorker: [https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web "OpenAI’s chatbot offers paraphrases, whereas Google offers quotes. Which do we prefer?"] To be fair, the version called Copilot by Microsoft does provide links to the original documents.

==Ethics==
Most of the effort in protecting the human population from the dangers of [[Artificial Intelligence]] focus on arguments about ethics. [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Quandaries in AI-ML is one of a growing list on YouTube] joined by an ACM webinar on [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Challenges in AI]. What remains unclear is how AI ethics can be separated from [[Censorship]]. Most of the effort at training AI involves a list of taboo subjects. The problem here is that not everyone agrees on what ethics should apply. For example, is abortion in one country to be considered to be murder and in another country just an exercise in free will. It is unclear whether a discussion of abortion methods, suicide or similar health issues is an ethical or a moral discussion. Similar issues are easy to list, such as free speech or social credit. It comes down to a question of whether ethics in AI will ever be any more that just censorship that is inconsistently applied?

===In Healthcare===
* [https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Forange.hosting.lsoft.com%2Ftrk%2Fclick%3Fref%3Dznwrbbrs9_6-2be2bx32c520x081583%26&data=04%7C01%7C%7C1b0cfc8a7db9484bb50308d946e0a1e9%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637618752869025374%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PaRyaYQadjyHbH5wP%2FruYlnD0B5rftISxc8a9dfAJeQ%3D&reserved=0 WHO Releases AI Guidelines for Health Government Technology] 2021-07-12 <blockquote>A new report from the World Health Organization (WHO) offers guidance for the ethical use of artificial intelligence (AI) in the health sector. The six primary principles for the use of AI as set forth in the report are to protect autonomy; promote human well-being, safety, and the public interest; ensure transparency, explainability, and intelligibility; foster [[Responsibility]] and accountability; ensure inclusiveness and equity; and promote responsive and sustainable AI. These principles are intended as a foundation for AI stakeholders, including governments, developers, and society. The report, Ethics and Governance of Artificial Intelligence for Health, finds public- and private-sector collaboration necessary to ensure AI’s accountability.<blockquote>

===Back to the Future===
Brian Christian described "The Alignment Problem"<ref>Brian Christian ''The Alignment Problem'' W. W. Norton (2020-10-06) ISBN 978-0-393-86833-3</ref> as the difficulty that AI has with aligning the results provided by the AI with society's goals. It is worse than you might imagine. The results from AI as of 2020 have not proven to be appropriate. This is, in part, because they are trained on data from social systems that are also deemed to be inappropriate. What most researchers are doing from this point is trying to determine how natual intelligence develops and feeding that back into the AI systems.

The first person to try to understand "The Construction of [[Reality]] in the Child' was Jean Piaget<ref>Jean Piaget ''The Construction of Reality in the Child'' (original 1954) Ballentine ISBN 345-02338-2-165</ref>. The first line of the book is "To understand how the budding intelligence constructs the external world, we must ask whether the child, in its first months of life, conceives and perceives things as we do, as object that have substance, that are permanent and of constant dimensions." Now that NMR brain scanners can image a thought within an infant's mind<ref>Rachel Fritts, ''How do Babies perceive the World?'' MIT News July/August 2021 p 24 ff</ref>, we can actually imagine that the brain really is just the Chinese Room that so perplexed Searle.<ref>James Somers, ''Head Space'' The New Yorker Dec. 6, 2021, p 30 ff.</ref> Searle's objection to [[Artificial Intelligence]] was that it was just a symbol manipulation scheme and that cannot be all that is involved in understanding. Interestingly at the same time Noam Chomsky created a generative linguistic model of language that was nothing more than the manipulation of symbols.<ref>George Lakof, ''The Functionalist's Dilemma'', American Scientist https://www.americanscientist.org/article/the-functionalists-dilemma</ref> As you might imagine, they disagreed often in print.<ref>''Chomsky’s Revolution: An Exchange'' (2002-07-18) https://www.nybooks.com/articles/2002/07/18/chomskys-revolution-an-exchange/</ref> See the wiki page on [[Knowledge]] for more on this dispute. Andy Clark brings together ideas and techniques from robotics, neuroscience, infant psychology, and artificial intelligence. He addresses a broad range of adaptive behaviors, from cockroach locomotion to the role of linguistic artifacts in higher-level thought.<ref>Andy Clark, ''Being There: Putting Brain, Body, and World Together Again'' 1996 https://direct.mit.edu/books/book/3917/Being-TherePutting-Brain-Body-and-World-Together</ref>

And finally, the AI research community is recognizing the turn of AI towards Natural Science and is not too happy about that?<ref>Subbarao Kambhampati, ''AI as (an Ersatz) Natural Science.'' '''CACM 65''' No 9 (2022-09) pp 8-9</ref> Kambhampati reports that "Indeed, in a 2003 lecture, Turing laureate Lesle Lamport sounded alarms about the very possibility of the future of computing belong to biology rather than logic, saying it will lead us to living in a world of homeopathy and faith healing<ref>Leslie Lamport, ''The Future of Computing: Logic or Biology'' (2003-07-23) https://bit.ly/3ahrsaI</ref> Perhaps we should go back to one of the authors of the ultimate book on logic, the Principia Mathematica Alfred North Whitehead<ref>Alfred North Whitehead, ''Principia Mathematica'' Cambridge UP (Second edition 1927)</ref> who said "Metaphysical categories are not dogmatic statements of the obvious, they are tentative formulation of the ultimate generalities. If we consider any scheme of philosophic categories as one complex assertion, and apply it to the logician's alternative, true or false, the answer must be that the scheme is false. The same answer must be given to like question respecting the existing formulate principles of any science."<ref>Alfred North Whitehead, ''Process and Reality'' Free Press (1929)</ref>

==Truth from the AI==
<pre>
Lying requires belief + intention.
Deception requires a goal to induce false belief.
AI has neither. So by definition, it can’t be “lying” or “deceiving” - at least not in the human sense.

Yet the desire to interpret internal model states as belief-like is so strong, researchers are now building probes to "extract" them.

Words like “belief,” “reasoning,” “understanding,” "thought", and “intelligence” are being borrowed from human cognition - without being earned, and devaluing them all in the process.

Not just by journalists, or AI researchers, but especially the ones hyping the tech.

Why?

Not just to replace workers and cut wages - they know if we have no money, we can’t consume. The system collapses.

This is about reshaping access to knowledge itself. Controlling what you see. How you learn. Who you become. Who you obey. Who you trust.

</pre>

===Effective Altruism===
Compare this with wiki page on [[Expected Value]].

Under the concept that those with wealth should be careful that the resources that the contribute back to society have a positive impact. In other words, they want to control the way their money is used. An effort to impact AI in this way is underway. <ref>Timnit Gerbu ''Effective Altruism Is Pushing a Dangerous Brand of "AI Safety"'' Wired (2022-11-30) https://www.wired.com/story/effective-altruism-artificial-intelligence-sam-bankman-fried/</ref><blockquote>EA is defined by the Center for Effective Altruism as “an intellectual project, using evidence and reason to figure out how to benefit others as much as possible.” And “evidence and reason” have led many EAs to conclude that the most pressing problem in the world is preventing an apocalypse where an artificially generally intelligent being (AGI) created by humans exterminates us. To prevent this apocalypse, Some of the billionaires who have committed significant funds to this goal include Elon Musk, Vitalik Buterin, Ben Delo, Jaan Tallinn, Peter Thiel, Dustin Muskovitz, and Sam Bankman-Fried, who was one of EA’s largest funders until the recent bankruptcy of his FTX cryptocurrency platform. As a result, all of this money has shaped the field of AI and its priorities in ways that harm people in marginalized groups while purporting to work on “beneficial artificial general intelligence” that will bring techno utopia for humanity. This is yet another example of how our technological future is not a linear march toward progress but one that is determined by those who have the money and influence to control it. </blockquote>

Nietzshe had a very dismal view on the efficacy of altruism<ref>Friedrich Nietzsche. ''The Twilight of the Idols'' no 35</ref> <blockquote>An "altruistic" morality, a morality under which selfishness withers, is in all circumstances a bad sign. This is true of individuals and above all of nations. The best are lacking when selfishness begins to be lacking. Instinctively to select that which is harmful to one, to be lured by "disinterested' motives,—these things almost provide the formula for decadence. "Not to have one's own interests at heart" —this is simply a moral fig-leaf concealing a very different fact, a physiological one, to no longer know how to find what is to my interest."... Disintegration of the instincts!—AII is up with man when he becomes altruistic.— instead of saying ingenuously "I am no longer any good," the lie Of morality in the decadent's mouth says: "Nothing is any good,—life is no good."—A judgment of this kind ultimately becomes a great danger; for it is infectious, and it soon flourishes on the soil of society...</blockquote>

A better answer to AI might be to allow those who work directly with underserved people. The best way to measure the success of a society is to see how it treats the least of its members.<ref>Karen Hao ''A new vision of artificial intelligence for the people'' Technology Review (2022-04-22)</ref><blockquote>This story is the fourth and final part of MIT Technology Review’s series on AI colonialism, the idea that artificial intelligence is creating a new colonial world order. The project is a radical departure from the way the AI industry typically operates. Over the last decade, AI researchers have pushed the field to new limits with the dogma “More is more”: Amass more data to produce bigger models (algorithms trained on said data) to produce better results. The approach has led to remarkable breakthroughs—but to costs as well. Companies have relentlessly mined people for their faces, voices, and behaviors to enrich bottom lines. And models built by averaging data from entire populations have sidelined minority and marginalized communities even as they are disproportionately subjected to the technology’s impacts. Over the years, a growing chorus of experts have argued that these impacts are '''repeating the patterns of colonial history.''' Global AI development, they say, is impoverishing communities and countries that don’t have a say in its development—the same communities and countries already impoverished by former colonial empires.</blockquote>

See the wiki [[Social Justice]] for more details include Herbert Simon on Bounded Rationality.

==Regulation==
The big question is it safe? The simple answer is that new technology always comes with new risks. Antitrust law is the best-known method for controlling enterprises too large to be controlled by the government. Perhaps they can be used to control a super intelligent being of another sort. Perhaps that's all an AI ever is, corporations with computers and imposed limitations.
* 2024-10-24 [https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/ Memorandum on Advancing the United States’ Leadership in Artificial Intelligence; Harnessing Artificial Intelligence to Fulfill National Security Objectives; and Fostering the Safety, Security, and Trustworthiness of Artificial Intelligence]
* NIST issued a call for participation on [https://www.federalregister.gov/documents/2023/11/02/2023-24216/artificial-intelligence-safety-institute-consortium Artificial Intelligence Safety Institute Consortium] 2023-11-02 in the federal resister.
* President Bidin issued a [https://www.whitehouse.gov/ostp/ai-bill-of-rights/ Blueprint for an AI Bill of Rights] (2022-10-04) problems are well documented. In America and around the world, systems supposed to help with patient care have proven unsafe, ineffective, or biased. Algorithms used in hiring and credit decisions have been found to reflect and reproduce existing unwanted inequities or embed new harmful bias and discrimination. Unchecked social media data collection has been used to threaten people’s opportunities, undermine their privacy, or pervasively track their activity—often without their knowledge or consent. Five Principles:
# Safe and Effective Systems - You should be protected from unsafe or ineffective systems.
# Algorithmic Discrimination Protections - You should not face discrimination by algorithms and systems should be used and designed in an equitable way.
# Data Privacy - You should be protected from abusive data practices via built-in protections and you should have [[Agency]] over how data about you is used.
# Notice and Explanation- You should know that an automated system is being used and understand how and why it contributes to outcomes that impact you.
# Human Alternatives, Consideration, and Fallback - You should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter.
* A law that went into effect in New York City in January 2023<ref>Richard Vanderford, ''New York AI Bias Law Prompts Uncertainty'' Wall Street Journal (2022-09-21) </ref> requires companies to audit their artificial intelligence (AI) hiring systems to assess whether they incorporate racial and gender biases. The law holds hiring companies liable for any biases and could impose fines for violations, but lacks clear guidelines for the AI audit process. While the city’s Department of Consumer and Worker Protection has not offered a timeline for when it will publish rules to implement the law, some companies already are taking steps to comply. Said Anthony Habayeb of AI governance software company Monitaur Inc., "Instead of waiting for someone to tell me what to do… I built controls around these applications because I know like with any software, things can and do go wrong."

==References==

<references />

===Other Material===
* See the wiki page on [[Intelligent Goals]].
* See the wiki page on [[Machine Learning]].
* See the wiki page on [[Fairness Accountability Transparency Ethics]]
* See the wiki page on [[Genetic Programming]]

[[Category: Glossary]]
[[Category: Model]]
[[Category: Authentication]]
[[Category: Artificial Intelligence]]

Artificial Intelligence

2026-03-23T00:32:42Z

Tom: /* Paradox */

==Full Title or Meme==
Intelligence without human lived experience or human compassion.

"The future is already here — it's just not very evenly distributed" - William Gibson.

The term [[Artificial Intelligence]] was coined, in part, out of spite. In 1955, the Dartmouth professor John McCarthy assembled a proposal for what became the first-ever conference on AI. McCarthy refused to use the prevailing term, "cybernetics": That would have necessitated inviting Norbert Wiener, then the world's leading cyberneticist, whom he found insufferable. Instead, McCarthy minted a new term that took off in the digital era, taking him along as a founder of the field. <ref>''Fun fact: Cybernetics'' Technology Review (2025-08) p 11</ref>

==Context==
===The Broad concepts===
* There are two great forces both in physics and in human history
** The harnessing of energy which started with the use of animal power to human benefit.
** The harnessing of information which started with philosophy and then Gutenberg's press.
* Likewise there are two great revolutions in human society.
** The Industrial revolution focused on control and use of energy which peaked in the 20th century.
** The Information revolution focused on control and use of information which is peaking in the 21st century.
* The social changes to accompany these revolution had to before the revolutions could complete.
** The educational system needed to be extended to all who work to engineer the control of energy.
** Now we need a second shift to educate most of the workers to control the information needed.
* The use of energy and information in harmony with the reality of the Earth's resources is a challenge.
** The profligate use of energy has altered the environment in ways that threaten human existence.
** The profligate use of information may prove to alter the environment of work to be a bigger threat.
* The struggle between conservative and progress forces in politics has been with us for ever.
** The conservative mind wants to hold onto the privilege of the past and present that they enjoy.
** The progressive mind want so shape the future to make that future a better place for us all.
* The Collingridge dilemma describes the challenge of regulating new technologies.
** Early stage: When a technology is new, its future impact is unclear, making regulation difficult.
** Later stage: Once the technology is widely adopted, it becomes harder to control or modify.

===Why Bother?===
Some jobs are too risky.
Some tasks are too mind-numbing.
Some conditions are too toxic.
Some decisions are too complex.

AI should be our answer to protect, preserve, and uplift human potential, not disregard or displace it.

===What is Intelligence===
* Here Intelligence relates primarily to [[Identity]] [[Knowledge]], but that has many general aspects.
* In 1950 Turing asked "Can Machines Think".<ref>A. M. Turing, ''COMPUTING MACHINERY AND INTELLIGENCE'' 1950-10-01 https://academic.oup.com/mind/article/LIX/236/433/986238</ref>
* People have been working on artificial intelligence starting with a conference at Dartmouth College in the 1950s. There, computer scientists and mathematicians set a very ambitious goal: to develop computers that could understand language, translate it, see patterns, and more.
* Marvin Minsky was teaching undergrad courses in 1964 using the book "Computers and Thought."<ref>Edward Feigenbaum, & Julian Feldman, joint ed ''Computers and Thought'' https://archive.org/details/computersthought00feig</ref>
* in 2023 Bill Gates in "AI is about to completely change how you use computers" said<ref>Bill Gates ''AI is about to completely change how you use computers'' https://www.gatesnotes.com/AI-agents</ref> "I still love software as much today as I did when Paul Allen and I started Microsoft. But—even though it has improved a lot in the decades since then—in many ways, software is still pretty dumb.:
* Some people think that [[Artificial Intelligence]] must needs be cold or sterile, but we have plenty of [[Evidence]] that is not so.
** Elisa was the first but [https://arstechnica.com/information-technology/2023/12/real-humans-appeared-human-63-of-the-time-in-recent-turing-test-ai-study/ it best chat gpt so I guess the turning test isn't so useful]
** Alexa works on Amazon echo devices and is constantly updated so there is nothing stable to compare.
** GPT was getting some press with GPT-2 included in Technology Review tagged it as a break-through technology in 2021<ref>Will Douglas, '' GPT-3'' Technology Review '''Vol 124''' No 2 (2021-03) pp. 34-35</ref>, but nothing managed to get public [[Attention]] until... {See below for more on Open AI}
** Chat-GPT which was the first publicly accessible product that was not completely cringe-worthy on 2022-11-30.
* Is [[Artificial Intelligence]] even [https://www.linkedin.com/pulse/machine-intelligence-artificial-sean-manion-phd-oz91e/ artificial]? Or is it just yet another way to create a balance between [[Chaos and Order]].
* Determining what [[Artificial Intelligence]] actually is and whether is is like human intelligence turns out to be a impossible task. Ray Kurzweil said<ref>Ray Kurzweil, ''The Singularity is Nearer''</ref><blockquote>The reason a well-defined empirical test is necessary is that ... humans have a powerful tendency to redefine whatever artificial intelligence achieves is not ''really'' so hard in hindsight. This is often referred to as the "AI effect". ... Intuitively [the hallucinations and odd answers] feels like a problem. It's tempting to think that Watson "ought" to reason like humans do. But i would propose that this is superstition. In the real world, what matters is how an intelligent being acts.</blockquote>
===Origin===
Over the summer of 1956 a small but illustrious group gathered at Dartmouth College in New Hampshire; it included Claude Shannon, the begetter of information theory, and Herbert A. Simon, the only person ever to win both the Nobel Memorial Prize in Economic Sciences awarded by the Royal Swedish Academy of Sciences and the Turing Award awarded by the Association for Computing Machinery. They had been called together by a young researcher, John McCarthy, who wanted to discuss “how to make machines use language, form abstractions and concepts” and “solve kinds of problems now reserved for humans”. It was the first academic gathering devoted to what McCarthy dubbed “artificial intelligence”. And it set a template for the field’s next 60-odd years in coming up with no advances on a par with its ambitions.<ref>John McCarthy, M. Minsky, N. Rochester, C.E Shannon, ''A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence'' (1955-08) http://raysolomonoff.com/dartmouth/boxa/dart564props.pdf</ref>

During 1955 and 1956 computer scientist and cognitive psychologist Allen Newell, political scientist, economist and sociologist Herbert A. Simon, and systems programmer John Clifford Shaw, all working at the Rand Corporation in Santa Monica, California, developed the Logic Theorist, the first program deliberately engineered to mimic the problem solving skills of a human being. They decided to write a program that could prove theorems in the propositional calculus like those in Principia Mathematica by Alfred North Whitehead and Bertrand Russell.<ref>''Newell, Simon & Shaw Develop the First Artificial Intelligence Program'' https://www.historyofinformation.com/detail.php?id=742</ref>

In 1956 John von Neumann agreed to present a discussion on "The Computer and the Brain" as the Yale Silliman lecture. He was too ill to complete the full gamut of lectures but did get enough presented to be published by Yale.<ref>John von Neumann, ''The Computer and the Brain'' (1958) Yale UP ISBN 0300024150</ref> Perhaps the most interesting part is the discussion of the difference between arithmetic and logic. He noted that only a few bits of information were available to make a choice between options, but that the computer's arithmetic was growing to every more bits, now 64 in most machines, to give sufficient precision to make a choice. Only machines defined after 2020 used ultra-low precision (2–4 bits) primarily for inference, not training. IBM researchers presented techniques for 4-bit quantization at NeurIPS 2020, focusing on reducing memory and compute costs while maintaining accuracy for deep neural networks

The first international conference was "Mechanisation of Thought Processes"<ref>National Physical Laboratory at Teddington,''Mechanisation of Thought Processes'' https://aitopics.org/download/aiclassics:98875BCD</ref> was held in 1958 with presentations by Minsky and John McCarthy.

===Perceptron===
Also called a McCulloch-Pitts [[Neuron]] is an attempt to mimic an animal [[Neuron]], originally with a purpose built computer system. While it was first described in 1947, it was not realized until Frank Rosenblatt built is Mark I computer using perceptrons as a primary physical element. He continued to build larger machines with more layer until his early death in 1971. Minskey and Papert wrote an entire book<ref> M. Minsky & S. Papert, ''Perceptrons'' (1987-12-28) ISBN 978-0262631112</ref> where they purported to show that it was a dead-end idea which basically blocked progress on them til the 1980's. The Large Language Models that became popular in the early 2020's are based on Multi-Level Perceptrons (MLP).

===End Game===
Now i am become death, the destroyer of worlds. Quoted by Robert Oppenheimer at the completion of the Atomic bomb and in the movie "Ex Machina" from the Bhagavad Gita.

Perhaps a good question to ask is, "Can Intelligence Be Separated from the Body?"<ref>Oliver Whang, ''Can Intelligence Be Separated From the Body?'' New York Times (2023-04-11) https://www.nytimes.com/2023/04/11/science/artificial-intelligence-body-robots.html?</ref><blockquote>Maybe the mind is like a video game controller, moving the body around the world, taking it on joy rides. Or maybe the body manipulates the mind with hunger, sleepiness and anxiety, something like a river steering a canoe. Is the mind like electromagnetic waves, flickering in and out of our light-bulb bodies? Or is the mind a car on the road? A ghost in the machine?</blockquote>

Perhaps a better question to ask is, "Can Intelligence Be Separated from a persistent Identifier"?<blockquote>The [[Identifier]]s used with AI in 2023 are basically just service [[Identifier]]s. in 2023 at the beginning of the general availability of the deployments of AIs that did not embarrass their creators, was also the time that context was provided to the AI to help it understand a thread of a conversation. Before that every query was de novo, without context. But still there is not a persistent history of conversations, or any concept that that is an "I" there to establish a history, which is still the basis of trust. Perhaps we must wait until we are able to crate an AI that persists an [[Identity]], we, as individuals, will be able to establish trust with it, as an individual. </blockquote>

==Problems==
* Humans seem to not be able to use [[Intelligent Design]] alone to fashion [[Artificial Intelligence]]. In fact researchers continue to go back to nature and human functioning for inspiration and even for algorithms. In other words, our understanding of the meaning of the term [[Artificial Intelligence]] has been and will continue to be evolving over time.
* Nature itself since the Big Bang seems to be based on [[Self-organization]] for its framework.
* Training an [[Artificial Intelligence]] with human behaviors will result in unacceptable behavior by the [[Artificial Intelligence]]. The most common complaint is real, or imagined, or invented, bias against one group or another.
** Microsoft released Tay,<ref>Peter Lee, Learning from Tay’s introduction. (2016-05-25) https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/</ref> a web robot (bot) to respond to tweets and chats<ref>Sarah Parez, ''Microsoft silences its new A.I. bot Tay, after Twitter users teach it racism.'' (2018-03-24) Tech Crunch https://techcrunch.com/2016/03/24/microsoft-silences-its-new-a-i-bot-tay-after-twitter-users-teach-it-racism/</ref> in May 2016. The result was disastrous as Tay learned to be racist from its all-too-human trainers and was shut down in days.
** Google has been plagued with reports and legal action on its search results nearly continuously since it was introduced; in 2019 the president, Donald Trump, accusing it of favoritism to leftist causes.<ref>Farhad Manjoo, ''Search Bias, Blind Spots And Google.'' (2018-08-31) New York Times p. B1</ref> Researcher Safiya U. Noble has written a book<ref>Safiya U. Noble, ''Algorithms of Oppression: How Search Engines Reinforce Racism.'' (2018-02-20) ISBN 978-1479849949</ref> mostly complaining that all-too-human programmers injected their own prejudices into their work. What else could he expect of humans, to rise above them selves, whatever that might mean in terms of free speech or freedom of religion?
** The page [[Right to be Forgotten]] describes an effort in Europe to teach the search engines to with-hold information about people that they don't like by collecting all of the information that they don't like. Be careful what you ask your [[Artificial Intelligence]] to do for you; it might just spill the beans some day, perhaps under court order.
** In the movie "Blade Runner 2049" the protagonist's AI girl friend asks to have her memory wiped so that the police cannot compel her to testify against him. One would hope that our future AIs will be that compassionate towards us; but that kind of compassionate behavior will probably be illegal.
===Personhood===
This is a really bad idea. It's solving a problem from the last generation of wallets when we need something that will work when the wallet is just an AI agent that can mimic anything thrown at it. Face it - the cred even today comes from a wallet that needs to prove its own creds before it can say anything about who is (or is not) at the controls. Face it - the internet of the future is just ai agents talking to each other.
* The development of full [[Artificial Intelligence]] could spell the end of the human race. - Stephan Hawking BBC Interview 2014. There are lots of dystopian futures depicted by science fiction authors. Some, like the Terminator series seem unlikely. The scenario in the movie Gattaca where humans were separated into categories based on the genome has already been enabled in the Personhood credential<ref>Pymnts, ''AI’s Human Mimicry Spurs ‘Personhood Credential’ Proposal'' 2024-09-24 https://www.pymnts.com/digital-identity/2024/ais-human-mimicry-spurs-personhood-credential-proposal/</ref> which would allow machines to determine by themselves who is qualified to be classified as human.
* Descartes claimed "I think therefore I am". But it seems that we need to be careful about applying that idea to [[Artificial Intelligence]]. Although his philosophy is no longer considered to be valid even for humans.
* [https://workshop.vennfactory.com/p/personhood-the-killer-credential Personhood: The Killer Credential?] Digging deeper into a very human challenge - Eve Maler 2024-09-16

===Vulnerabilities===
* 2024-08-20 [https://www.w3.org/reports/ai-web-impact/#safety-and-security AI & the Web: Understanding and managing the impact of Machine Learning models on the Web] from the W3C
* [https://home.treasury.gov/news/press-releases/jy2212U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector] U.S. Department of the Treasury 2024-03-27
* [https://arxiv.org/pdf/2402.00888.pdf Security and Privacy Challenges of Large Language Models: A Survey] (2024-02)
* Input to the model used by the AI come from two sources, each of which can be compromised.
# Training data from the Internet can be tampered with to give bad results.
## For starters, the internet data is typically censored to remove input that would cause embarrassment and [[Conduct Risk]] for the sponsors.
## Some locales will censor data to make themselves look good or promote some social point of view of the government of of the industry.
# The input from a single user can cause unexpected changes to later results.
## The early AI release were compromised with a variety of hate speech that quickly resulted in really nasty output from the AI.
## Artists seeking to protect their intellectual property from later output have learned to put small changes in the images that cause them to be mislabeled or worse.<ref>Melissa Hakkiliia, ''This new data-poisoning tool lets artists fight back against generative AI'' MIT Technology Review 127 No 1 (2024-01) p 7=8</ref>
* Giving a AI a goal may result in unexpected and undesired solutions.
** Bruce Schneier described<ref>Bruce Schneier, ''Hackers Used to Be Humans. Soon, AIs Will Hack Humanity'' Wired (2021-04-18) https://www.wired.com/story/opinion-hackers-used-to-be-humans-soon-ais-will-hack-humanity/?bxid=5c5b250d24c17c67f8640083</ref> AI to be Like crafty genies. They will grant our wishes and then hack them, exploiting our social, political, and economic systems like never before.

See the wiki on [[AI Threat Model]]s for more details on this subject.

===Uniquely Human===
Neither the Turning Test, nor any other known test, is designed to determine if an [[Artificial Intelligence]] is really human. The best way to understand what natural intelligence is would be to look at the reason human evolved to have such a large brain. There appear to be these principle requirements:<ref>Edward O. Wilson, ''The Meaning of Human Existences'' Liveright (2014) Chapter 2 ISBN 9780871401007</ref>
# Necessity for a large brain is the changing [[Ecosystem]] that a small remnant of Homo Sapiens living in the south of Africa needed in order to survive
# Changing locations to track new sources of protein and the switch from plant to animal protein.
# Empathy to be able to understand what is in someone else's mind to deduce if they notice you and if they think you are targeted by them as prey.
The results of these were two processes that served to enable survival in such varying [[Ecosystem]]s/
# Passing learning from one generation to the next.
# An extended adolescence with mothers that ceased being fecund to tend to their existing children and the extended life of grandparents to help with stories and other techniques to bring the knowledge of the tribe to the growing child.
The search for causes that were more complex than the ability of genes to carry information, such as the migratory pattern of birds is described with more detail in the wiki page on [[Chance and Necessity]].

* [https://www.linkedin.com/pulse/machine-intelligence-artificial-part-4-cybernetics-sean-manion-phd-kfi0e Machine Intelligence is not Artificial - Part 4: Cybernetics and Norbert Wiener]<blockquote>What is of particular note is not simply the existence of cybernetics as a discipline dating back to the 1940s, but the broad popularity Wiener's book and the topic received not only from scientists and engineers, but also the general public. His 1948 book went through four printings in the first 6 months and was a best seller, rivaling the contemporary Kinsey Report in popularity over the following decade. This was despite this book, ostensibly for a lay audience, being extremely dense with many chapters full of mathematical equations. One of the chapters in the original version, "The Computing Machines and the Nervous System," makes it clear that Wiener's mention of humans AND machines in the book's subtitle was pointing towards the comparison of these systems, making this book one of the early popularizers of the idea of "thinking machines" (but by no means the only one). This was two years before Alan Turing's 1950 paper "Computing Machinery and Intelligence," which gave us the Turing test for AI.</blockquote>

* Human-centered AI - [https://aiindex.stanford.edu/report/?utm_source=Stanford+HAI&utm_campaign=525788e5b7-AI_INDEX_2024_EMAIL_CAMPAIGN&utm_medium=email&utm_term=0_-525788e5b7-%5BLIST_EMAIL_ID%5D&mc_cid=525788e5b7&mc_eid=e55581d782 THE AI INDEX REPORT - Measuring trends in AI] 2024 HAI Stanford

===Explaining AI===
Most users want to know '''WHY''' a decision was make or an action was performed. The article "Solving for Why"<ref> Marina Kraskovsky, ''Solving for Why'' '''CACM 65''' No 2 (2202-02) pp 11ff</ref> is a focus on [[Causality]] to help researchers overcome shortcomings that have bedeviled more traditional approaches to AI. It is not enough to predict things, a good AI will need to solves things and explain why the choice was make if the decision is to have any support by the human population. People will ask if the decision was fair and will second guess the choices make by the AI unless they understand and accept the decision. Today AI decision are fragile, a slight change in the environment may have drastic changes in outcomes. This may well be inexplicable to humans. It will be insufficient to convince experts in the field about the choices made. It is apparent in 2022 that expert opinions about COVID have not been very popular. Even more evidence will be required of a AI decision. This will be particularly true when the decision is costly in any way. Humans like to have choices, which may require the AI to offer choices. That will have the benefit of giving the humans some feeling that they are in control. But beware of "Hobson's choice". That was a stable master at Oxford that was required to give the students a choice of horses, but selected the options so that there was no real choice.

===Creativity===
Jasmine Sun is currently The Atlantic’s clearest voice arguing that AI degrades creativity by eliminating stakes, judgment, and responsibility—even as it perfects surface‑level fluency.<ref>Sasmine Sun, ''The Human Skill That Eludes AI'' The Atlantic 2026-03-17</ref>
===Paradox===
Moravec’s Paradox describes our experience that it is hard to teach machines to do things that are easy for most humans (walking, for example) but comparatively easy to teach them things that are challenging for most humans (chess comes to mind).

Polanyi’s Paradox is named for the philosopher Michael Polanyi (1891-1976), who developed the concept of “tacit knowledge”: Central to Michael Polanyi’s thinking was the belief that creative acts (especially acts of discovery) are shot-through or charged with strong personal feelings and commitments (hence the title of his most famous work Personal Knowledge). Arguing against the then dominant position that science was somehow value-free, Michael Polanyi sought to bring into creative tension a concern with reasoned and critical interrogation with other, more ‘tacit’, forms of knowing. Polanyi’s argument was that the informed guesses, hunches and imaginings that are part of exploratory acts are motivated by what he describes as ‘passions’. They might well be aimed at discovering ‘truth’, but they are not necessarily [[Formal Model]]s that can be stated in propositional or formal terms. As Michael Polanyi (1967: 4) wrote in The Tacit Dimension, we should start from the fact that ‘we can know more than we can tell‘.<ref>MARK K. SMITH, “MICHAEL POLANYI AND TACIT KNOWLEDGE” AT INFED</ref>

* For synonyms to tacit knowledge see the wiki page on [[Common Sense]].
* It seems, in 2023, that generative AI is able to overcome the lack of imagination with simple random guessing where knowledge is lacking.

===Fairness===
* What is called "Algorithmic Bias" today is just the result of training any [[Artificial Intelligence]] using any data based on human behavior which typically is biased.
* We now expect more ''Fairness'' from our [[Artificial Intelligence]] as we expect from normal humans.
* "The burgeoning field of algorithmic fairness, part of the much broader field of responsible computing, is aiming to remedy the situation"<ref> Marina Krakovsky, '' Formalizing Fairness'' '''CACM 65''' No. 8 (2022-08) p 11ff.</ref> which is the result of using available data.
* Scientists have found that the English word "fairness" has many meanings, and do they would prefer terms like technical disparities, that they can define precisely. The problem with that is that what society demands is "fairness". Which means that scientists could never deliver what society demands.
* As was demonstrated in the COMPASS program to predict recidivism in convicts, it is not possible create a single solution that satisfies more than one metric for "fairness".
* Explainable Fairness means can you explain it to a judge so that they can make a determination. -- Cathy O'Neil ORCAA
* Every provider wants you to think that they are meeting the presidents EO, like the guys pitching Zero Trust Network Access (ZTNA) - do not be confused, this meets no level of [[Least Privilege]] or any other attempt to limit access to people that have earned access. It is nothing but the network folk trying to climb on band wagon.
* [https://www.readynez.com/en/blog/understanding-fair-ai-a-simple-guide/ Understanding Fair AI: A Simple Guide!]
* See the wiki on [[Fairness Accountability Transparency Ethics]] for more details on this topic.

===Disruption===
We can now say what GPT can replace - shallow human conversation and mechanical activity like law or medicine clerks perform. A GPT in a cocktail party could be wittier than anyone else. A GPT could do better than a recent grad of most law or medical schools at citing existing law or medical practice. Is anyone surprised? AI is moving up the food chain from one mechanical task to the next. We learned from aircraft pilot errors that we had to have a checklist to save pilots from error. I guess we could call the current AI a check list of stuff that is already known.

The question about what is left for humans might again be of interest. It was during the industrial revolution when machines replaced human labor. The best that the pundits of the time could suggest it that humans could have more leisure time. Now machines are moving to more tasks. But somehow employment is full in 2022. There is likely to be a reconning some day when machines and their owners will disrupt the system and cause unemployment. The rich and the poor will, yet again, come into conflict and a revolution of some sort will be required.

===Bias===
* Clearly any [[Artificial Intelligence]] is biased by its input data. Social pressures have forced any [[Entity]] that offers an AI to make if conform to the laws and regulations of the state. All countries have proscribed some set of topics or responses as unacceptable. In other words all AI are censored by the training data or the laws of the land. Even the language used for training can create biases.<ref>Antony Chayka and Andrei Sukhov, ''Comparing Chatbots Trained in Different Languages'' '''CACM 66''' no 12 pp, 6-7 (2023-12)</ref><blockquote>In our opinion, the government's position is clearly taken into account in the responses of AI systems in the national language, especially when the creation of AI was funded in the tested country.</blockquote>
* We face the same problems in raising our children. If they are raised in an environment that exhibits bias, it will be hard to change their behavior later in life.<ref>Will Douglas Heaven, ''Six Big Questions for Generative AI: Hope, Hype, Nope'' Technology Review (2024-01) p. 30ff. </ref> But if we use synthetic data to train the AI there will be huge objections as it is not possible for any training material (for either humans or AI) to be objectionable to some subset of humanity.

==Solutions==
The wiki page on [[Trust]] describes how people can get into the habit of trusting business process or technology process that really do not deserve to be trusted. The following is a listing of some of the technological solutions. The question about whether an [[Artificial Intelligence]] is trustworthy is yet to be addressed.

===Levels===
It is to be expected that level of intelligence will be applicable, which the higher levels learning from, or dependent on, simpler levels that start from basic automation and then to the following.<ref>Peter Denning and Ted Lewis, ''An AI Learning Hierarchy'' '''CACM 67''' No 12 (2024-12) p. 24ff.</ref>

# Rule-based system, these were applied to board games early in the computing age, but were not able to Handle more complex games like chess or go where experts were unable to articulate the rules by which they played.
# Supervised Learning
# Unsupervised Learning
# Generative AI
# Reinforcement Learning
# Human-machine interaction
# Aspirational AI

===Governance===
For most of 2023 the focus was on government standards for AI. The idea that trust can be initiated by government oversight is wide-spread, but wrong. Trust comes from a history of fail dealing, that is not something that governments understand or practice. But still the government can, and should, punish crimes, so the legislation should focus on how crimes committed by AI are punished. The death penalty would be one good deterrent. Here is one discussion on what governments can do:
[https://www.linkedin.com/pulse/governance-trust-guy-huntington-cus3c/ AI/bots and National Security].
===Training===
A great deal of effort is directed at the education of young children. In large part education is directed at the production of adults that a useful the the society in which they are expected to function as useful citizens. But their is a sharp disagreement on the subject. According to Ken Robinson, education should have four core purposes: personal, economic, social, and cultural <ref>Ken Robinson, Kate Robinson ''What Is Education For?'' (2022-03-02) https://www.edutopia.org/article/what-education</ref> Perhaps it is time to model the training of AI on the training of humans?

===User Experience===
The real question about the use of the current Large Language Model of [[Artificial Intelligence]] is how the device in our possession displays the interaction with the [[Artificial Intelligence]]. We humans have been reading stories and watching displays of [[Artificial Intelligence]] from the puppet Pinocchio that wanted to be a real boy to Deus ex Machina a movie about an [[Artificial Intelligence]] robot that went rouge. The common thread here is that each was a fictional account that we were able to relate to. Good authors know how to enliven the characters of a story for the reader that drags the reader into the story. The [[User Experience]] of AI for the near term will be interactions through our personal devices, even though the more interesting, and troubling, case will be when the robotic devices are not in our possession, but directly run by some gigantic corporation. The good story tellers will be the ones that grab our [[Attention]], even as is true today.<ref>Hannah H. Kim, ''A.I. is a Fiction'' Wired pp. 9-10 (2023-08) 31-08</ref>

What we must focus on is the UX and its effect on the people that interact with it.

===The Smart Agent===
Now that an A.I. can apply reason to problems, it should be expected that such agents on on the horizon. That prediction was made in 2021, and there seems to be glimmerings of AI appearing in the user agents in our hands in 2023, but it is still all driven by Large Language Models (LLM) in the cloud. Intel is predicting that user devices in 2024 will include and AI. Loading LLM into [[Edge Computing]] devices are needed before AI will work in small devices.

Nvidia has been building GPUs for personal computers for a long time before they were purposed for [[Artificial Intelligence]]. But on 2023-12-14 they were not used for that on personal computers when Intel unveiled [https://www.intel.com/content/www/us/en/newsroom/news/ai-everywhere-core-ultra-5th-gen-xeon-news.html#gs.26279f "Core Ultra and 5th Gen Xeon – that will change how customers and consumers enable AI solutions in the data center, cloud, network, at the edge, and on the PC"] to bring AI to any home computer user with a budget for a new computer.

===Neural Nets===

Still the most popular, they appear to have reached a limit in what they can explain about human intelligence. Perhaps we need to add compassion and [[Common Sense]] back in.

The theory of neural nets has been documented in Quanta <ref>Kevin Hartnett, ''Foundations Built for a General Theory of Neural Networks.'' (2019-01-31) Quanta https://www.quantamagazine.org/foundations-built-for-a-general-theory-of-neural-networks-20190131/</ref> A description of the Turing award to the winners for applying neural nets to solving problem (as opposed to explaining intelligence) was summarized by Neil Savage<ref>Neil Savage, ''Neural Net Worth.'' '''CACM Vol 62''' no 6 p. 12 (2019-06)</ref> who specifically talks about the need for more involvement of humanities in long term solutions.

===Deep Neural Network Accelerators===
Now we come to the logical and nearly the final phase of [[Artificial Intelligence]], the design of one AI by another.<ref>Adam Zewe, ''Accelerating AI tasks while preserving data security'' MIT News (2023-10-30) https://news.mit.edu/2023/accelerating-ai-tasks-while-preserving-data-security-1030</ref> This is probably just the first step to the complete control of the AI [[Ecosystem]] by the AI themselves when the designs become to complex for human understanding.<blockquote>With the proliferation of computationally intensive machine-learning applications, such as chatbots that perform real-time language translation, device manufacturers often incorporate specialized hardware components to rapidly move and process the massive amounts of data these systems demand. Choosing the best design for these components, known as deep neural network accelerators, is challenging because they can have an enormous range of design options. This difficult problem becomes even thornier when a designer seeks to add cryptographic operations to keep data safe from attackers. Now, MIT researchers have developed a search engine that can efficiently identify optimal designs for deep neural network accelerators, that preserve data security while boosting performance.</blockquote>

===Libratus===

Libratus did not use expert domain knowledge, or human data that are specific to [the game domain]. Rather, the AI was able to analyze the game's rules and devise its own strategy. The technology thus could be applied to any number of imperfect-information games. Such hidden information is ubiquitous in real-world strategic interactions, including business negotiation, cybersecurity, finance, strategic pricing and military applications.<ref>Byron Spice, ''CSD's Sandholm, Brown To Receive Minsky Medal.'' (2018-11-06) Carnegie-Mellon News https://www.cmu.edu/news/stories/archives/2018/november/minsky-medal.html</ref>

===Alpha Zero===

After Alpha Go, the first version to use [[Matrix Calculation|Tensor Processing Units]] (TPU) defeating the best human GO play in 2016,<ref>McMillan, Robert. "Google Isn't Playing Games With New Chip". (2016-05-18) The Wall Street Journal. Archived from the original on 29 June 2016. </ref> the Google Deep Mind project Alpha Go was generalized into a far simpler version Alpha Zero<ref>Steven Strogatz, ''One Giant Step for a Chess-Playing Machine.'' (2018-12-26) New York Times https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html</ref> that was just given the rules of the game, whether Go or Chess, and allowed to play against another instance of itself until it had mastered the subject to be able to beat any existing computer model. Garry Kasparov, the world chess said that Alpha Go developed a style of play that "reflects the truth" of the game rather than "the priorities and prejudices of programmers." That pretty much ends the discussion about the superiority of "Intelligent Design" which turns out to be just another myth that humans tell about themselves and their anthropomorphic gods. It also ended the testing of [[Artificial Intelligence]] against humans in board games. There was very little generally newsworthy announcements about [[Artificial Intelligence]] for the next 6 years.

===Open AI===
Open AI Is a research institute founded by Elon Musk and Sam Altman to make AI results available to everyone. Their finding (which echoed that of Deep Mind at Google) in early 2019 was that this is a very bad idea. Like any technology it can ready be turned to evil purposes as reported in Wired.<ref>Alyssa Foote, ''The AI Text Generator That's Too Dangerous to Make Public.'' (219-02-19) Wired Magazine https://www.wired.com/story/ai-text-generator-too-dangerous-to-make-public/</ref> This finding should certain give pause to any company that seeks to use AI for their own purposes. It is too easy for AI to be turned to the purposed of evil intent. Sam Altman wrote an essay ''Moore's Law for Everything''<ref>Sam Altman ''Moore's Law for Everything'' personal blog (2021) https://moores.samaltman.com/</ref> that was reviewed by Ezra Klein.<ref>Ezra Klein ''Sam Altman on the A.I. Revolution, Trillionaires and the Future of Political Power'' NY Time (2021-06-11) https://www.nytimes.com/2021/06/11/opinion/ezra-klein-podcast-sam-altman.html?action=click&module=Opinion&pgtype=Homepage</ref>

Not everyone is captivated by Open AI. Quote from the New Yorker: [https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web "OpenAI’s chatbot offers paraphrases, whereas Google offers quotes. Which do we prefer?"] To be fair, the version called Copilot by Microsoft does provide links to the original documents.

==Ethics==
Most of the effort in protecting the human population from the dangers of [[Artificial Intelligence]] focus on arguments about ethics. [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Quandaries in AI-ML is one of a growing list on YouTube] joined by an ACM webinar on [https://www.youtube.com/results?search_query=Ethical+Quandaries+in+AI-ML Ethical Challenges in AI]. What remains unclear is how AI ethics can be separated from [[Censorship]]. Most of the effort at training AI involves a list of taboo subjects. The problem here is that not everyone agrees on what ethics should apply. For example, is abortion in one country to be considered to be murder and in another country just an exercise in free will. It is unclear whether a discussion of abortion methods, suicide or similar health issues is an ethical or a moral discussion. Similar issues are easy to list, such as free speech or social credit. It comes down to a question of whether ethics in AI will ever be any more that just censorship that is inconsistently applied?

===In Healthcare===
* [https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Forange.hosting.lsoft.com%2Ftrk%2Fclick%3Fref%3Dznwrbbrs9_6-2be2bx32c520x081583%26&data=04%7C01%7C%7C1b0cfc8a7db9484bb50308d946e0a1e9%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637618752869025374%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=PaRyaYQadjyHbH5wP%2FruYlnD0B5rftISxc8a9dfAJeQ%3D&reserved=0 WHO Releases AI Guidelines for Health Government Technology] 2021-07-12 <blockquote>A new report from the World Health Organization (WHO) offers guidance for the ethical use of artificial intelligence (AI) in the health sector. The six primary principles for the use of AI as set forth in the report are to protect autonomy; promote human well-being, safety, and the public interest; ensure transparency, explainability, and intelligibility; foster [[Responsibility]] and accountability; ensure inclusiveness and equity; and promote responsive and sustainable AI. These principles are intended as a foundation for AI stakeholders, including governments, developers, and society. The report, Ethics and Governance of Artificial Intelligence for Health, finds public- and private-sector collaboration necessary to ensure AI’s accountability.<blockquote>

===Back to the Future===
Brian Christian described "The Alignment Problem"<ref>Brian Christian ''The Alignment Problem'' W. W. Norton (2020-10-06) ISBN 978-0-393-86833-3</ref> as the difficulty that AI has with aligning the results provided by the AI with society's goals. It is worse than you might imagine. The results from AI as of 2020 have not proven to be appropriate. This is, in part, because they are trained on data from social systems that are also deemed to be inappropriate. What most researchers are doing from this point is trying to determine how natual intelligence develops and feeding that back into the AI systems.

The first person to try to understand "The Construction of [[Reality]] in the Child' was Jean Piaget<ref>Jean Piaget ''The Construction of Reality in the Child'' (original 1954) Ballentine ISBN 345-02338-2-165</ref>. The first line of the book is "To understand how the budding intelligence constructs the external world, we must ask whether the child, in its first months of life, conceives and perceives things as we do, as object that have substance, that are permanent and of constant dimensions." Now that NMR brain scanners can image a thought within an infant's mind<ref>Rachel Fritts, ''How do Babies perceive the World?'' MIT News July/August 2021 p 24 ff</ref>, we can actually imagine that the brain really is just the Chinese Room that so perplexed Searle.<ref>James Somers, ''Head Space'' The New Yorker Dec. 6, 2021, p 30 ff.</ref> Searle's objection to [[Artificial Intelligence]] was that it was just a symbol manipulation scheme and that cannot be all that is involved in understanding. Interestingly at the same time Noam Chomsky created a generative linguistic model of language that was nothing more than the manipulation of symbols.<ref>George Lakof, ''The Functionalist's Dilemma'', American Scientist https://www.americanscientist.org/article/the-functionalists-dilemma</ref> As you might imagine, they disagreed often in print.<ref>''Chomsky’s Revolution: An Exchange'' (2002-07-18) https://www.nybooks.com/articles/2002/07/18/chomskys-revolution-an-exchange/</ref> See the wiki page on [[Knowledge]] for more on this dispute. Andy Clark brings together ideas and techniques from robotics, neuroscience, infant psychology, and artificial intelligence. He addresses a broad range of adaptive behaviors, from cockroach locomotion to the role of linguistic artifacts in higher-level thought.<ref>Andy Clark, ''Being There: Putting Brain, Body, and World Together Again'' 1996 https://direct.mit.edu/books/book/3917/Being-TherePutting-Brain-Body-and-World-Together</ref>

And finally, the AI research community is recognizing the turn of AI towards Natural Science and is not too happy about that?<ref>Subbarao Kambhampati, ''AI as (an Ersatz) Natural Science.'' '''CACM 65''' No 9 (2022-09) pp 8-9</ref> Kambhampati reports that "Indeed, in a 2003 lecture, Turing laureate Lesle Lamport sounded alarms about the very possibility of the future of computing belong to biology rather than logic, saying it will lead us to living in a world of homeopathy and faith healing<ref>Leslie Lamport, ''The Future of Computing: Logic or Biology'' (2003-07-23) https://bit.ly/3ahrsaI</ref> Perhaps we should go back to one of the authors of the ultimate book on logic, the Principia Mathematica Alfred North Whitehead<ref>Alfred North Whitehead, ''Principia Mathematica'' Cambridge UP (Second edition 1927)</ref> who said "Metaphysical categories are not dogmatic statements of the obvious, they are tentative formulation of the ultimate generalities. If we consider any scheme of philosophic categories as one complex assertion, and apply it to the logician's alternative, true or false, the answer must be that the scheme is false. The same answer must be given to like question respecting the existing formulate principles of any science."<ref>Alfred North Whitehead, ''Process and Reality'' Free Press (1929)</ref>

==Truth from the AI==
<pre>
Lying requires belief + intention.
Deception requires a goal to induce false belief.
AI has neither. So by definition, it can’t be “lying” or “deceiving” - at least not in the human sense.

Yet the desire to interpret internal model states as belief-like is so strong, researchers are now building probes to "extract" them.

Words like “belief,” “reasoning,” “understanding,” "thought", and “intelligence” are being borrowed from human cognition - without being earned, and devaluing them all in the process.

Not just by journalists, or AI researchers, but especially the ones hyping the tech.

Why?

Not just to replace workers and cut wages - they know if we have no money, we can’t consume. The system collapses.

This is about reshaping access to knowledge itself. Controlling what you see. How you learn. Who you become. Who you obey. Who you trust.

</pre>

===Effective Altruism===
Compare this with wiki page on [[Expected Value]].

Under the concept that those with wealth should be careful that the resources that the contribute back to society have a positive impact. In other words, they want to control the way their money is used. An effort to impact AI in this way is underway. <ref>Timnit Gerbu ''Effective Altruism Is Pushing a Dangerous Brand of "AI Safety"'' Wired (2022-11-30) https://www.wired.com/story/effective-altruism-artificial-intelligence-sam-bankman-fried/</ref><blockquote>EA is defined by the Center for Effective Altruism as “an intellectual project, using evidence and reason to figure out how to benefit others as much as possible.” And “evidence and reason” have led many EAs to conclude that the most pressing problem in the world is preventing an apocalypse where an artificially generally intelligent being (AGI) created by humans exterminates us. To prevent this apocalypse, Some of the billionaires who have committed significant funds to this goal include Elon Musk, Vitalik Buterin, Ben Delo, Jaan Tallinn, Peter Thiel, Dustin Muskovitz, and Sam Bankman-Fried, who was one of EA’s largest funders until the recent bankruptcy of his FTX cryptocurrency platform. As a result, all of this money has shaped the field of AI and its priorities in ways that harm people in marginalized groups while purporting to work on “beneficial artificial general intelligence” that will bring techno utopia for humanity. This is yet another example of how our technological future is not a linear march toward progress but one that is determined by those who have the money and influence to control it. </blockquote>

Nietzshe had a very dismal view on the efficacy of altruism<ref>Friedrich Nietzsche. ''The Twilight of the Idols'' no 35</ref> <blockquote>An "altruistic" morality, a morality under which selfishness withers, is in all circumstances a bad sign. This is true of individuals and above all of nations. The best are lacking when selfishness begins to be lacking. Instinctively to select that which is harmful to one, to be lured by "disinterested' motives,—these things almost provide the formula for decadence. "Not to have one's own interests at heart" —this is simply a moral fig-leaf concealing a very different fact, a physiological one, to no longer know how to find what is to my interest."... Disintegration of the instincts!—AII is up with man when he becomes altruistic.— instead of saying ingenuously "I am no longer any good," the lie Of morality in the decadent's mouth says: "Nothing is any good,—life is no good."—A judgment of this kind ultimately becomes a great danger; for it is infectious, and it soon flourishes on the soil of society...</blockquote>

A better answer to AI might be to allow those who work directly with underserved people. The best way to measure the success of a society is to see how it treats the least of its members.<ref>Karen Hao ''A new vision of artificial intelligence for the people'' Technology Review (2022-04-22)</ref><blockquote>This story is the fourth and final part of MIT Technology Review’s series on AI colonialism, the idea that artificial intelligence is creating a new colonial world order. The project is a radical departure from the way the AI industry typically operates. Over the last decade, AI researchers have pushed the field to new limits with the dogma “More is more”: Amass more data to produce bigger models (algorithms trained on said data) to produce better results. The approach has led to remarkable breakthroughs—but to costs as well. Companies have relentlessly mined people for their faces, voices, and behaviors to enrich bottom lines. And models built by averaging data from entire populations have sidelined minority and marginalized communities even as they are disproportionately subjected to the technology’s impacts. Over the years, a growing chorus of experts have argued that these impacts are '''repeating the patterns of colonial history.''' Global AI development, they say, is impoverishing communities and countries that don’t have a say in its development—the same communities and countries already impoverished by former colonial empires.</blockquote>

See the wiki [[Social Justice]] for more details include Herbert Simon on Bounded Rationality.

==Regulation==
The big question is it safe? The simple answer is that new technology always comes with new risks. Antitrust law is the best-known method for controlling enterprises too large to be controlled by the government. Perhaps they can be used to control a super intelligent being of another sort. Perhaps that's all an AI ever is, corporations with computers and imposed limitations.
* 2024-10-24 [https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/ Memorandum on Advancing the United States’ Leadership in Artificial Intelligence; Harnessing Artificial Intelligence to Fulfill National Security Objectives; and Fostering the Safety, Security, and Trustworthiness of Artificial Intelligence]
* NIST issued a call for participation on [https://www.federalregister.gov/documents/2023/11/02/2023-24216/artificial-intelligence-safety-institute-consortium Artificial Intelligence Safety Institute Consortium] 2023-11-02 in the federal resister.
* President Bidin issued a [https://www.whitehouse.gov/ostp/ai-bill-of-rights/ Blueprint for an AI Bill of Rights] (2022-10-04) problems are well documented. In America and around the world, systems supposed to help with patient care have proven unsafe, ineffective, or biased. Algorithms used in hiring and credit decisions have been found to reflect and reproduce existing unwanted inequities or embed new harmful bias and discrimination. Unchecked social media data collection has been used to threaten people’s opportunities, undermine their privacy, or pervasively track their activity—often without their knowledge or consent. Five Principles:
# Safe and Effective Systems - You should be protected from unsafe or ineffective systems.
# Algorithmic Discrimination Protections - You should not face discrimination by algorithms and systems should be used and designed in an equitable way.
# Data Privacy - You should be protected from abusive data practices via built-in protections and you should have [[Agency]] over how data about you is used.
# Notice and Explanation- You should know that an automated system is being used and understand how and why it contributes to outcomes that impact you.
# Human Alternatives, Consideration, and Fallback - You should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter.
* A law that went into effect in New York City in January 2023<ref>Richard Vanderford, ''New York AI Bias Law Prompts Uncertainty'' Wall Street Journal (2022-09-21) </ref> requires companies to audit their artificial intelligence (AI) hiring systems to assess whether they incorporate racial and gender biases. The law holds hiring companies liable for any biases and could impose fines for violations, but lacks clear guidelines for the AI audit process. While the city’s Department of Consumer and Worker Protection has not offered a timeline for when it will publish rules to implement the law, some companies already are taking steps to comply. Said Anthony Habayeb of AI governance software company Monitaur Inc., "Instead of waiting for someone to tell me what to do… I built controls around these applications because I know like with any software, things can and do go wrong."

==References==

<references />

===Other Material===
* See the wiki page on [[Intelligent Goals]].
* See the wiki page on [[Machine Learning]].
* See the wiki page on [[Fairness Accountability Transparency Ethics]]
* See the wiki page on [[Genetic Programming]]

[[Category: Glossary]]
[[Category: Model]]
[[Category: Authentication]]
[[Category: Artificial Intelligence]]

Creative Destruction

2026-03-23T00:23:55Z

Tom: /* Others on Disruption */

==Full Title or Meme==
Static societies do not innovate. Chaotic societies do not survive. A measured approach to [[Creative Destruction]]<ref name="schumpeter">Joseph Schumpeter ''Capitalism, Socialism and Democracy'' Routledge; (1942)
ISBN 978-1138138612</ref> can build dynamic, growing societies.

==Context==
Corporate reformers strongly preach Joseph Schumpeter<ref name="schumpeter" /> and Clayton Christensen's theory<ref>Clayton M. Christensen, ''The Innovator’s Dilemma''. (1997-01-05) Harvard Business Review Press ISBN 978-1633691780</ref> of “creative destruction,” “innovation, and “disruption” as driving corporate history. These ideas justify their efforts to tear apart traditional public school systems, replace experienced teachers with inexperienced youngsters, close schools, and experiment with charters, vouchers, and anything else that will destroy the status quo. To be sure, some are in the school reform business to make money, but others see themselves as heroes of a movement that sees itself as blowing up “failing schools” and forcing fresh innovations into a stagnant sector of the economy.

This author used Christensen's argument in a pitch to Bill Gates for his "think week". Gates blasted the paper for sloppy thinking. His objections were not fully understood until a remarkable article by Jill Lepore, published by the New Yorker, exploded the dogmas of “disruption” as progress. Do schools need to be disrupted by techniques borrowed from the business world? Do families need to be disrupted? Do communities need disruption? According to disruption theory, disruption is the precursor to success.

==Lepore disrupts Christensen==
Lepore was at Harvard with the business professor Clayton Christensen. He popularized the idea that big companies die as they are overtaken by nimble start-ups that embrace innovation. Business leaders took heed and committed themselves to persistent re-invention and self-disruption, or buying up the start-ups before they overtook the established industry leader.

In education, we have seen the dogma of disruption in the policies of Arne Duncan, the Bloomberg administration of education in New York City (with its focus on closing schools and opening schools and closing the schools it opened), the Rahm Emanuel model (closing 50 public schools on the same day), the Broad Foundation, the Gates Foundation, the Walton Foundation, and business groups. They scorn incremental change and pursue the disruptive idea–like closing schools, the Common Core, federally funded tests–that will shake up schools across the nation with a series of bold and experimental strokes.

Lepore, it can be fairly said, demolishes disruption theory by showing that Christensen’s examples provide no evidence for the theory. To the contrary, the successful companies over the long haul were not the innovators that disrupted the industry, but those that changed incrementally, tinkering and constantly improving their processes and their products. One of Christensen’s leading examples of disruption was the disk-drive industry. This was the subject of his doctoral dissertation. In his telling, a company called Seagate Technology fell by the wayside as competitors disrupted its market. But, Lepore shows, Christensen was wrong.<ref>Jill Lepore, ''The Disruption Machine'' (2014-06-16) The New Yorker. https://www.newyorker.com/magazine/2014/06/23/the-disruption-machine </ref>
<blockquote>In fact, Seagate Technology was not felled by disruption. Between 1989 and 1990, its sales doubled, reaching $2.4 billion, “more than all of its U.S. competitors combined,” according to an industry report. In 1997, the year Christensen published “The Innovator’s Dilemma,” Seagate was the largest company in the disk-drive industry, reporting revenues of nine billion dollars. Last year, Seagate shipped its two-billionth disk drive. Most of the entrant firms celebrated by Christensen as triumphant disrupters, on the other hand, no longer exist, their success having been in some cases brief and in others illusory….. As striking as the disruption in the disk-drive industry seemed in the nineteen-eighties, more striking, from the vantage of history, are the continuities. Christensen argues that incumbents in the disk-drive industry were regularly destroyed by newcomers. But today, after much consolidation, the divisions that dominate the industry are divisions that led the market in the nineteen-eighties. (In some instances, what shifted was their ownership: I.B.M. sold its hard-disk division to Hitachi, which later sold its division to Western Digital.) In the longer term, victory in the disk-drive industry appears to have gone to the manufacturers that were good at incremental improvements, whether or not they were the first to market the disruptive new format. Companies that were quick to release a new product but not skilled at tinkering have tended to flame out.</blockquote>
Lepore continues:
<blockquote>Christensen’s sources are often dubious and his logic questionable. His single citation for his investigation of the “disruptive transition from mechanical to electronic motor controls,” in which he identifies the Allen-Bradley Company as triumphing over four rivals, is a book called “The Bradley Legacy,” an account published by a foundation established by the company’s founders. This is akin to calling an actor the greatest talent in a generation after interviewing his publicist. “Use theory to help guide data collection,” Christensen advises.</blockquote>
Lepore continues:
<blockquote>Disruptive innovation as an explanation for how change happens is everywhere. Ideas that come from business schools are exceptionally well marketed. Faith in disruption is the best illustration, and the worst case, of a larger historical transformation having to do with secularization, and what happens when the invisible hand replaces the hand of God as explanation and justification. Innovation and disruption are ideas that originated in the arena of business but which have since been applied to arenas whose values and goals are remote from the values and goals of business. People aren’t disk drives. Public schools, colleges and universities, churches, museums, and many hospitals, all of which have been subjected to disruptive innovation, have revenues and expenses and infrastructures, but they aren’t industries in the same way that manufacturers of hard-disk drives or truck engines or drygoods are industries. Journalism isn’t an industry in that sense, either.</blockquote>
Note that the hard disk drive was still manufactured by Seagate (after several more mergers) in 2025 even though disruptions continue and solid state drives (SSD) are offered for many of the markets that were dominated by the HDD when the article was written.
<blockquote>Doctors have obligations to their patients, teachers to their students, pastors to their congregations, curators to the public, and journalists to their readers—obligations that lie outside the realm of earnings, and are fundamentally different from the obligations that a business executive has to employees, partners, and investors. Historically, institutions like museums, hospitals, schools, and universities have been supported by patronage, donations made by individuals or funding from church or state. The press has generally supported itself by charging subscribers and selling advertising. (Underwriting by corporations and foundations is a funding source of more recent vintage.) Charging for admission, membership, subscriptions and, for some, earning profits are similarities these institutions have with businesses. Still, that doesn’t make them industries, which turn things into commodities and sell them for gain.</blockquote>

==Others on Disruption==
<blockquote>When Alexander Graham Bell introduced his telephone in March 1876, the invention was riddled with problems. The line was a crackly mess—prone to interference from nearby electrical lines—and it was powered by a battery that leaked acid. Still, it allowed for a remarkable, unmooring experience: For the first time, you could talk in real time to someone blocks or miles away. “It was like a voice from another world,” marveled one early user. Bell quickly improved the quality, and customers thronged. In the first year, over 3,000 telephones sold; by 1900 there were over one million phones nationwide.

At first, the telephone was marketed mainly as a tool for business. Physicians and drugstores bought them to process orders, and business owners installed them at home so they could be quickly reached. The phone, proclaimed early ad copy, gave business leaders an ESP-like “sixth sense” of their far-flung operations.

The idea of using such a powerful tool for everyday conversation? That seemed laughable and obnoxious. One early social critic warned that the phone should not be used for the “exchange of twaddle between foolish women.” Businessmen prohibited their wives from tying up the line, lest they interfere with commerce. “At the beginning, women were forbidden to use the phone—the business was supposed to have the priority,” notes Michéle Martin, a professor emeritus at Canada’s Carleton University and author of Hello, Central?

But it quickly became apparent that people wanted to talk—to socialize. One phone company manager in 1909 did a survey of usage and found that 30 percent of all calls were “idle gossip,” lasting 7.5 minutes on average each. He didn’t like this chitchat, but he was running against the stream. Eventually phone firms realized there was more money in selling lines for banter than for business. “They realized, ‘We can make money off gossip and idle conversation and sociability on the telephone,’” says Claude Fischer, author of America Calling: A Social History of the Telephone to 1940.</blockquote>
Clive Thompson ''Texting Isn’t the First New Technology Thought to Impair Social Skills When Alexander Graham Bell introduced the telephone, skeptics worried about how it might affect people’s interactions'' Smithsonian Magazine 2016-03 https://www.smithsonianmag.com/innovation/texting-isnt-first-new-technology-thought-impair-social-skills-180958091/
===Capitalism===
Sven Beckert’s work shows capitalism is not conservative, either in the political‑ideological sense or in the sense of defending the status quo. The sources are quite clear on this, and we can say so without speculating about his personal politics, simply by sticking to what he explicitly argues and how his work is received.<ref>Sven Beckert ''Capitalism'' 2025</ref>

==References==
<references />
===Other Material===
* [https://economics.mit.edu/files/1785#:~:text=run%20macroeconomic%20consequences.-,Creative%20destruction%20refers%20to%20the%20incessant%20product%20and%20process%20innovation,the%20essential%20fact%20about%20capitalism' MIT course work] on [[Creative Destruction]].
* The thread of this wiki follows [https://dianeravitch.net/2015/03/11/jill-lepore-the-hoax-of-disruptive-innovation/ Dian Ravitch's blog] (2015-03-11).
* [https://www.publicbooks.org/public-thinker-jill-lepore-on-the-challenge-of-explaining-things/ Interview with Jill Lepore] on creative destruction and publishing.

[[Category: Glossary]]
[[Category: Article]]

Policy Language

2026-03-22T23:24:23Z

Tom: /* Existing Languages */

==Full Title or Meme==
A description of the syntax and semantics of a machine-readable [[Language]] to enable access decisions.

==Context==
There are two broad areas of policy:
# Business rules
# Government Legislation and Regulation (which can include regulations that are situationally dependent like health or emergency.)
*One area that is not covered here is moral rules of expected behavior. That is addressed in the page on [[Conduct Risk]].
* Also see the wiki page on [[Policy-Based Access Control]].

==Existing Languages==
Originally these languages were all written by geeks, and for geeks. What is needed is some policy that works in a real-world user journey. In other words, the user needs to be able to understand what is expected and when. It might be that a [[Policy Language]] can be a meta language that can be translated into commonly used language so that the user can understand the impact of the policy, or it might be that the [[Policy Language]] is only used to move the policy from one machine to another, but even then debugging would be easier if the [[Policy Language]] could be rendered into something that a human could understand.
===Natural Language Programming===
Full paper. It's a short one. https://www.cs.utexas.edu/~EWD/transcriptions/EWD06xx/EWD667.html

On the foolishness of "natural language programming" . Since the early days of automatic computing we have had people that have felt it as a shortcoming that
===Formal Proofs===
Iteration is powerful, but it blows up the state space of a policy. Once you allow unbounded loops, it becomes computationally intractable — sometimes outright impossible — to exhaustively explore every possible execution path. That means you can’t reliably prove universal statements like:
*“No confidential records will ever be exposed to public users.”
* “This partner domain will never be able to write to this dataset.”
* “All requests will be decided within 50ms.”

With Cedar’s no-iteration model, those proofs are tractable. Because the evaluation graph is finite and schema-constrained, analyzers can enumerate all possible outcomes. That gives you deterministic complexity, portability across trust domains, and guarantees you can enforce in governance agreements.

Yes — it’s inconvenient at times. You might have to pre-process data in the calling app or refactor logic that would be a simple for loop in Rego. But you gain:

* Static verification: Certainty about safety properties without runtime tests.
* Predictable performance: Every decision has a known bound on time and resources.
* Schema enforcement: No surprises from mismatched data shapes.
* Interoperability: Other parties can confidently consume and trust your policies.

In interdomain security, that trade is worth it. An “expressive” policy you can’t prove correct is just a bug waiting for the right input. Cedar chooses analyzability over iteration so we can build trust that spans org boundaries — and in that world, provability beats cleverness every time.

===Cedar===
Cedar is a policy syntax invented by Amazon. It's used by the AWS Verified Permissions, Authz-as-a-Service offering. Gluu is working on a new product at the Janssen Project called the "Cedarling"--which leverages the Cedar policy syntax and Amazon's open source Cedar Rust engine. The Cedarling can run anywhere--as a local agent in the browser, embedded in a mobile application, or as a cloud service. It needs no data, because it trusts the JWT tokens that are input to the request by the application. Beyond policy evaluation, the Cedarlng agent has two other capabilities: JWT validation and audit logging. In this episode, Mike will present Gluu's current progress on the Cedarling and show a demo of the Cedarling in action!

Cedarling provides an app that creates its own instance ID. This is part of opensource Jensen project.

Cedar is a new authorization-policy language used by the Amazon Verified Permissions and AWS Verified Access managed services, and we recently released it publicly. Using Cedar, developers can write policies that specify fine-grained permissions for their applications. The applications then authorize access requests by calling Cedar’s authorization engine. Because Cedar policies are separate from application code, they can be independently authored, updated, analyzed, and audited. <ref>Mike Hicks ''How we built Cedar with automated reasoning and differential testing'' 2023-05-19 https://www.amazon.science/blog/how-we-built-cedar-with-automated-reasoning-and-differential-testing</ref>

* [https://www.cedarpolicy.com/en Cedar Policy Language website.]
* [https://www.linkedin.com/events/7254252166016614400/comments/ Cedarling Launch]
* [https://www.linkedin.com/events/7439439950414073857/ Episode 189: Agentic Authorization with Cedar] Event by Identerati Tue, May 5, 2026
news:
* 2025-08-21 [https://aws.amazon.com/about-aws/whats-new/2025/08/amazon-verified-permissions-cedar-4-5/ Amazon Verified Permissions now supports Cedar 4.5]
* 2025-06-17 [https://aws.amazon.com/blogs/opensource/introducing-cedar-analysis-open-source-tools-for-verifying-authorization-policies/ Introducing Cedar Analysis: Open Source Tools for Verifying Authorization Policies]

===Open Policy Agent===
[https://www.openpolicyagent.org/ OPA, the Open Policy Agent]
* 2025-08-20 [https://blog.openpolicyagent.org/note-from-teemu-tim-and-torin-to-the-open-policy-agent-community-2dbbfe494371 Note from Teemu, Tim, and Torin to the Open Policy Agent community] Merged into Apple
* Declarative. Express policy in a high-level, declarative [[Policy Language]] that promotes safe, performant, fine-grained controls. Use a language purpose-built for policy in a world where JSON is pervasive. Iterate, traverse hierarchies, and apply 150+ built-ins like string manipulation and JWT decoding to declare the policies you want enforced.
* Context-aware. Leverage external information to write the policies you really care about. Stop inventing roles that represent complex relationships that years down the road no one will understand. Instead, write logic that adapts to the world around it and attach that logic to the systems that need it.

===Rego===
Rego is a high‑level declarative policy language created for the Open Policy Agent (OPA). Inspired by Datalog, it’s designed to work with structured data (like JSON). Rego is extremely expressive. It’s almost Turing-complete, with the ability to iterate over arbitrary collections, perform recursion, and generate new data structures at runtime. This flexibility is a big reason OPA has been adopted widely in infrastructure-as-code validation, Kubernetes admission control, and API authorization. If you need to traverse complex nested JSON or write sophisticated matching logic, Rego can do it — and often in a very natural way for developers.
* Declarative: Focus on what conditions must hold, not how to enforce them.
* Structured data support: Works natively with JSON, making it ideal for cloud‑native systems.
* Composable rules: Policies can be broken into small, reusable rules.
* Portable: Same language and engine can be used across microservices, Kubernetes, CI/CD pipelines, APIs, and more.

https://www.openpolicyagent.org/docs/latest/policy-language/

https://medium.com/@bhavini_sehgal/exploring-role-based-access-control-rbac-and-rego-a-simplified-guide-to-policy-management-932db0fcdf97

===Compare Cedar to Rego===

**side‑by‑side code example** to see how the same policy idea (“only admins can delete resources”) looks in **Rego (OPA)** versus **Amazon Cedar**.
Example Policy: “Only users with the role `admin` can delete resources”

Rego (OPA)**
* rego
package example.authz
Input is JSON, e.g.:
{
"user": {"roles": ["admin"]},
"action": "delete",
"resource": "server1"
}

default allow = false

allow {
input.action == "delete"
"admin" in input.user.roles
}

* How it works:**
** OPA evaluates the `allow` rule.
** If the action is `"delete"` and the user has `"admin"` in their roles, the rule returns `true`.
** Otherwise, access is denied.
*Flexibility:** You can add more rules, combine attributes, or pull in external JSON data.
*Amazon Cedar**
cedar
// Define a policy template
permit(
principal in Role::"admin",
action == Action::"delete",
resource in Resource::"server"
);

*How it works:**
** Cedar policies are written in terms of **principals** (who), **actions** (what), and **resources** (on what).
** This policy permits any principal in the `admin` role to perform the `delete` action on resources of type `server`.
*Simplicity:** Cedar is purpose‑built for authorization, so the syntax is concise and domain‑specific.

Contrast
*Rego (OPA):**
** General‑purpose, works with arbitrary JSON data.
** More expressive but requires more boilerplate.
** Can be used for *any* policy domain (compliance, admission control, API auth, etc.).

*Cedar (AWS):**
** Purpose‑built for access control.
** Easier to read/write for authorization, but narrower scope.
** Tightly integrated with AWS Verified Permissions.

So: **Rego** is like a *Swiss Army knife* for policy enforcement across many domains, while **Cedar** is a *specialist scalpel* for access control in AWS.

===Cedarling===

Cedarling: An Embeddable Policy Decision Point

Cedarling is an open-source, embeddable Policy Decision Point (PDP) that runs anywhere—from the browser or mobile app to sidecars or cloud services—returning fine-grained authorization decisions based on **Amazon’s Cedar** policy syntax and the open-source Cedar Rust engine.
# Core Capabilities
*Policy Evaluation**: Converts incoming JWTs into Cedar entities (principal, action, resource, context) and evaluates them against declarative access policies using the PARC model (Policy, Action, Resource, Context).
**JWT Validation**: Trusts and validates tokens from configured issuers, ensuring only well-formed, unexpired, unrevoked tokens drive authorization decisions.
**Audit Logging**: Emits detailed decision logs—denied or allowed—with tokens, policies, contexts, and timestamps, feeding SIEM or ITDR platforms for real-time threat detection.
# Architecture & Modules == Cedarling’s Rust-based codebase is organized into key modules for flexibility and performance:
*authz**: Core policy evaluation logic
*entity_builder**: Maps JWT claims to Cedar entities
*jwt**: Signature, issuer, and expiration checks
*log**: Structured audit logging
*bootstrap_config & init**: Load and initialize configuration
*http**: Multi-target safe HTTP helpers for policy updates and revocation checks
*lock**: (Optional) Communicates with a “Lock Server” for dynamic policy pushes and telemetry
*tests**: End-to-end coverage ensuring correctness
# Performance & Use Cases
*Ultra-Low Latency**: Typically issues authz decisions in under 50 microseconds, making it suitable for high-volume, latency-sensitive environments.
*Legacy Instrumentation**: Can be injected into existing applications without altering business logic—evaluating requests and logging decisions risk-free.
*Cloud-Native & Edge**: Ideal for zero-trust scenarios, from API gateways and Kubernetes clusters to databases and IoT gateways.
# Ecosystem & Adoption
*Janssen Project**: Cedarling is a core PDP within the Linux Foundation’s Janssen open-source IAM suite (alongside OAuth/OIDC, SCIM, FIDO modules).
*Gluu Flex**: Commercial distributions embed Cedarling for enterprise deployments, integrating with OAuth status lists and centralized threat-detection workflows.
*Third-Party Integrations**: Used by platforms like StrongDM for zero-trust access and by SaaS vendors via Cedar-agent for standalone PDP functionality.

*In essence, Cedarling brings Amazon Cedar’s expressive, purpose-built authorization language into a lightweight, embeddable Rust library—enabling consistent, auditable access control at any layer of your architecture.**

===Graph Query Lanage===
https://graphql.org/

===Cardea===
They have been working with the government of Aruba and come up with the most through look at the problem in 2022. The have picked the unfortunate name of '''Machine Readable Governance''', but it really is a description of the government's policy. Unfortunately, it might change daily and does not meet the requirement of testing by the user and remaining valid for even the duration of a flight from the US to the island.
* [https://hackmd.io/@mikekebert/HyjBmusEF#/ Machine Readable Governance.] A JWT with a mishmash of different levels of abstraction from human readable questions to access policy.
* [https://github.com/thecardeaproject/cardea Cardea on GitHub.]

===XACML===
* [http://docs.oasis-open.org/xacml/3.0/xacml-3.0-core-spec-os-en.doc XACML 3.0 core spec.]
* This spec is written in XML with no regard of size of the message. XML can easily be converted to json and this is the spec that describes that process.
* While it is oriented to an overly complex structure using monolithic designs, it has well-thought-out elements that should remain useful in the domain where it was applied. It is less useful in new areas like application of areas like [[COVID Vaccination]].
* A lot of the language used below is taken from the XACML spec. For example:
** PDP = policy definition point
** PEP = policy execution point
===Cerbos===
Explore the emergence of a new class of WASM (WebAssembly) embeddable Policy Decision Points (PDP) products, such as Cerbos’ ePDP and the Linux Foundation Janssen Project Cedarling, and why they represent a significant leap forward for modern enterprises.

===LegalRuleML===
Is an [https://www.oasis-open.org/2021/09/28/oasis-open-approves-legalruleml-core-v1-0-standard-for-legislation-contract-and-case-law/ OASIS spec] that attempts to turn law into machine readable format. This wiki page is focused more on the application of the policy to access control.
* Converting law, written in English (in the US) to a policy to be applied for access control is insanely difficult.<ref>Esther Shein, '' Converting Laws to Programs'' pp 15-16 (2022-01) '''Communications of the ACM Vol. 65''' No 1 https://m.acmwebvm01.acm.org/magazines/2022/1/257436-converting-laws-to-programs/fulltext</ref> Typically the programmer will be given a design document that is coded and ship before the lawyers are asked to adjudicate some discrepancy. Trying to engage the lawyers at the start of the process might sound better, unless you were a program manager with a schedule to keep. And a policy is a basically a programmer's view of the policy in the spec. The likelihood of error is close to one.

* Adam Wyner commented:
It is very welcome to see a discussion in the ACM about artificial intelligence/computer science applied to law. Applications to tax go all the way back to 1977 with the Taxman system and subsequent work:
https://www.jstor.org/stable/1340132
Further work on AI and Law can be found on the International Association for Artificial Intelligence and Law:
http://www.iaail.org/
and the Journal of Artificial Intelligence and Law:
https://www.springer.com/journal/10506

===Drools===
* Supposedly this is open, which apparently applies to the java code implementation which is at the following site.
* [https://docs.jboss.org/drools/release/7.1.0.Final/drools-docs/html_single/ Drools documentation.]
* "Drools is Rule Engine or a Production Rule System that uses the rule-based approach to implement an [[Expert System]]. [[Expert System]]s are knowledge-based systems that use knowledge representation to process acquired knowledge into a knowledge base that can be used for reasoning."
* The rules supported by Drools look a lot like a domain specific programming language, not a business rules language.

===HL7 CQL===
* [https://cql.hl7.org Clinical Quality Language] (CQL) is a high-level, domain-specific language focused on clinical quality and targeted at measure and decision support artifact authors.
* In addition, this specification describes a machine-readable canonical representation called Expression Logical Model (ELM) targeted at implementations and designed to enable sharing of clinical knowledge.

===Microsoft ADMX Group Policy Language===
* [https://learn.microsoft.com/en-us/previous-versions/dotnet/articles/bb530196(v=msdn.10)?redirectedfrom=MSDN Managing Group Policy ADMX Files Step-by-Step Guide]

===PAM===
Privileged Access Management (PAM) is a cybersecurity strategy and set of technologies designed to control, monitor, and secure access to critical systems and data by users with elevated permissions—known as privileged users.

====Core Concepts====
Privileged Access: Special permissions that allow users to perform sensitive tasks—like configuring systems, accessing confidential data, or managing user accounts.

Privileged Accounts: These include admin accounts, root accounts, service accounts, and others that can bypass standard security controls.

====Key Functions of PAM====
Access Control: Ensures only authorized users can access privileged accounts.

Session Monitoring: Tracks and records activities during privileged sessions for audit and compliance.

Credential Management: Automates password rotation and vaulting to prevent misuse.

Just-in-Time Access: Grants temporary access only when needed, reducing exposure.

Least Privilege Enforcement: Limits users to the minimum access necessary to perform their tasks.==Current Activity==

===IETF===
The first attempt at [https://datatracker.ietf.org/doc/charter-ietf-policy/ A Policy Framework Charter] (2004-11-05) expired two decades ago.

===CAREDA===

==User Agent Example==
This section shows the requirements placed on a [[User Agent]] by some corresponding site, which could include sources like private data or user credentials. User agents are code provided by a trusted source that can act on behalf of the holder of the device that contains an instance of the [[User Agent]]. Note that some parts of the [[User Agent]] may be cloud based, but the device resident part will be able to function even if the cloud is inaccessible.
# A [[Trusted Execution Environment]] (TEE, TPM, Secure Enclave, etc.) may be required on the holder's device.
# Similar to a local firewall or the ICS (Internet Connection Sharing) on Windows a policy that either blocks or allows URLs.
# An internal firewall may limit access to specified site names or IP addresses depending on the data access rules.
# Protection of user private keys may be required to be protected in a TEE.
# Some classes of data may be required to be encrypted at rest by a key held the the TEE.
# Data in motion may require a TLS of a particular level (eg. 1.3) to share data.
# The screen capture functionality may be blocked to prevent data exfiltration.
# The user presence at the device may be required for zero access requests.
# Continued user liveness tests at the device may be required continuously or all access will be blocked.
# A collection of Trust Zones or Federations may be required to access data bound to them.

==References==
<references />
===Other Material===
* See wiki [[Policy-Based Access Control]]

[[Category: Authorization]]
[[Category: Policy]]
[[Category: Language]]

Policy Language

2026-03-22T23:17:24Z

Tom: /* Cedar */

==Full Title or Meme==
A description of the syntax and semantics of a machine-readable [[Language]] to enable access decisions.

==Context==
There are two broad areas of policy:
# Business rules
# Government Legislation and Regulation (which can include regulations that are situationally dependent like health or emergency.)
*One area that is not covered here is moral rules of expected behavior. That is addressed in the page on [[Conduct Risk]].
* Also see the wiki page on [[Policy-Based Access Control]].

==Existing Languages==
Originally these languages were all written by geeks, and for geeks. What is needed is some policy that works in a real-world user journey. In other words, the user needs to be able to understand what is expected and when. It might be that a [[Policy Language]] can be a meta language that can be translated into commonly used language so that the user can understand the impact of the policy, or it might be that the [[Policy Language]] is only used to move the policy from one machine to another, but even then debugging would be easier if the [[Policy Language]] could be rendered into something that a human could understand.
===Formal Proofs===
Iteration is powerful, but it blows up the state space of a policy. Once you allow unbounded loops, it becomes computationally intractable — sometimes outright impossible — to exhaustively explore every possible execution path. That means you can’t reliably prove universal statements like:
*“No confidential records will ever be exposed to public users.”
* “This partner domain will never be able to write to this dataset.”
* “All requests will be decided within 50ms.”

With Cedar’s no-iteration model, those proofs are tractable. Because the evaluation graph is finite and schema-constrained, analyzers can enumerate all possible outcomes. That gives you deterministic complexity, portability across trust domains, and guarantees you can enforce in governance agreements.

Yes — it’s inconvenient at times. You might have to pre-process data in the calling app or refactor logic that would be a simple for loop in Rego. But you gain:

* Static verification: Certainty about safety properties without runtime tests.
* Predictable performance: Every decision has a known bound on time and resources.
* Schema enforcement: No surprises from mismatched data shapes.
* Interoperability: Other parties can confidently consume and trust your policies.

In interdomain security, that trade is worth it. An “expressive” policy you can’t prove correct is just a bug waiting for the right input. Cedar chooses analyzability over iteration so we can build trust that spans org boundaries — and in that world, provability beats cleverness every time.

===Cedar===
Cedar is a policy syntax invented by Amazon. It's used by the AWS Verified Permissions, Authz-as-a-Service offering. Gluu is working on a new product at the Janssen Project called the "Cedarling"--which leverages the Cedar policy syntax and Amazon's open source Cedar Rust engine. The Cedarling can run anywhere--as a local agent in the browser, embedded in a mobile application, or as a cloud service. It needs no data, because it trusts the JWT tokens that are input to the request by the application. Beyond policy evaluation, the Cedarlng agent has two other capabilities: JWT validation and audit logging. In this episode, Mike will present Gluu's current progress on the Cedarling and show a demo of the Cedarling in action!

Cedarling provides an app that creates its own instance ID. This is part of opensource Jensen project.

Cedar is a new authorization-policy language used by the Amazon Verified Permissions and AWS Verified Access managed services, and we recently released it publicly. Using Cedar, developers can write policies that specify fine-grained permissions for their applications. The applications then authorize access requests by calling Cedar’s authorization engine. Because Cedar policies are separate from application code, they can be independently authored, updated, analyzed, and audited. <ref>Mike Hicks ''How we built Cedar with automated reasoning and differential testing'' 2023-05-19 https://www.amazon.science/blog/how-we-built-cedar-with-automated-reasoning-and-differential-testing</ref>

* [https://www.cedarpolicy.com/en Cedar Policy Language website.]
* [https://www.linkedin.com/events/7254252166016614400/comments/ Cedarling Launch]
* [https://www.linkedin.com/events/7439439950414073857/ Episode 189: Agentic Authorization with Cedar] Event by Identerati Tue, May 5, 2026
news:
* 2025-08-21 [https://aws.amazon.com/about-aws/whats-new/2025/08/amazon-verified-permissions-cedar-4-5/ Amazon Verified Permissions now supports Cedar 4.5]
* 2025-06-17 [https://aws.amazon.com/blogs/opensource/introducing-cedar-analysis-open-source-tools-for-verifying-authorization-policies/ Introducing Cedar Analysis: Open Source Tools for Verifying Authorization Policies]

===Open Policy Agent===
[https://www.openpolicyagent.org/ OPA, the Open Policy Agent]
* 2025-08-20 [https://blog.openpolicyagent.org/note-from-teemu-tim-and-torin-to-the-open-policy-agent-community-2dbbfe494371 Note from Teemu, Tim, and Torin to the Open Policy Agent community] Merged into Apple
* Declarative. Express policy in a high-level, declarative [[Policy Language]] that promotes safe, performant, fine-grained controls. Use a language purpose-built for policy in a world where JSON is pervasive. Iterate, traverse hierarchies, and apply 150+ built-ins like string manipulation and JWT decoding to declare the policies you want enforced.
* Context-aware. Leverage external information to write the policies you really care about. Stop inventing roles that represent complex relationships that years down the road no one will understand. Instead, write logic that adapts to the world around it and attach that logic to the systems that need it.

===Rego===
Rego is a high‑level declarative policy language created for the Open Policy Agent (OPA). Inspired by Datalog, it’s designed to work with structured data (like JSON). Rego is extremely expressive. It’s almost Turing-complete, with the ability to iterate over arbitrary collections, perform recursion, and generate new data structures at runtime. This flexibility is a big reason OPA has been adopted widely in infrastructure-as-code validation, Kubernetes admission control, and API authorization. If you need to traverse complex nested JSON or write sophisticated matching logic, Rego can do it — and often in a very natural way for developers.
* Declarative: Focus on what conditions must hold, not how to enforce them.
* Structured data support: Works natively with JSON, making it ideal for cloud‑native systems.
* Composable rules: Policies can be broken into small, reusable rules.
* Portable: Same language and engine can be used across microservices, Kubernetes, CI/CD pipelines, APIs, and more.

https://www.openpolicyagent.org/docs/latest/policy-language/

https://medium.com/@bhavini_sehgal/exploring-role-based-access-control-rbac-and-rego-a-simplified-guide-to-policy-management-932db0fcdf97

===Compare Cedar to Rego===

**side‑by‑side code example** to see how the same policy idea (“only admins can delete resources”) looks in **Rego (OPA)** versus **Amazon Cedar**.
Example Policy: “Only users with the role `admin` can delete resources”

Rego (OPA)**
* rego
package example.authz
Input is JSON, e.g.:
{
"user": {"roles": ["admin"]},
"action": "delete",
"resource": "server1"
}

default allow = false

allow {
input.action == "delete"
"admin" in input.user.roles
}

* How it works:**
** OPA evaluates the `allow` rule.
** If the action is `"delete"` and the user has `"admin"` in their roles, the rule returns `true`.
** Otherwise, access is denied.
*Flexibility:** You can add more rules, combine attributes, or pull in external JSON data.
*Amazon Cedar**
cedar
// Define a policy template
permit(
principal in Role::"admin",
action == Action::"delete",
resource in Resource::"server"
);

*How it works:**
** Cedar policies are written in terms of **principals** (who), **actions** (what), and **resources** (on what).
** This policy permits any principal in the `admin` role to perform the `delete` action on resources of type `server`.
*Simplicity:** Cedar is purpose‑built for authorization, so the syntax is concise and domain‑specific.

Contrast
*Rego (OPA):**
** General‑purpose, works with arbitrary JSON data.
** More expressive but requires more boilerplate.
** Can be used for *any* policy domain (compliance, admission control, API auth, etc.).

*Cedar (AWS):**
** Purpose‑built for access control.
** Easier to read/write for authorization, but narrower scope.
** Tightly integrated with AWS Verified Permissions.

So: **Rego** is like a *Swiss Army knife* for policy enforcement across many domains, while **Cedar** is a *specialist scalpel* for access control in AWS.

===Cedarling===

Cedarling: An Embeddable Policy Decision Point

Cedarling is an open-source, embeddable Policy Decision Point (PDP) that runs anywhere—from the browser or mobile app to sidecars or cloud services—returning fine-grained authorization decisions based on **Amazon’s Cedar** policy syntax and the open-source Cedar Rust engine.
# Core Capabilities
*Policy Evaluation**: Converts incoming JWTs into Cedar entities (principal, action, resource, context) and evaluates them against declarative access policies using the PARC model (Policy, Action, Resource, Context).
**JWT Validation**: Trusts and validates tokens from configured issuers, ensuring only well-formed, unexpired, unrevoked tokens drive authorization decisions.
**Audit Logging**: Emits detailed decision logs—denied or allowed—with tokens, policies, contexts, and timestamps, feeding SIEM or ITDR platforms for real-time threat detection.
# Architecture & Modules == Cedarling’s Rust-based codebase is organized into key modules for flexibility and performance:
*authz**: Core policy evaluation logic
*entity_builder**: Maps JWT claims to Cedar entities
*jwt**: Signature, issuer, and expiration checks
*log**: Structured audit logging
*bootstrap_config & init**: Load and initialize configuration
*http**: Multi-target safe HTTP helpers for policy updates and revocation checks
*lock**: (Optional) Communicates with a “Lock Server” for dynamic policy pushes and telemetry
*tests**: End-to-end coverage ensuring correctness
# Performance & Use Cases
*Ultra-Low Latency**: Typically issues authz decisions in under 50 microseconds, making it suitable for high-volume, latency-sensitive environments.
*Legacy Instrumentation**: Can be injected into existing applications without altering business logic—evaluating requests and logging decisions risk-free.
*Cloud-Native & Edge**: Ideal for zero-trust scenarios, from API gateways and Kubernetes clusters to databases and IoT gateways.
# Ecosystem & Adoption
*Janssen Project**: Cedarling is a core PDP within the Linux Foundation’s Janssen open-source IAM suite (alongside OAuth/OIDC, SCIM, FIDO modules).
*Gluu Flex**: Commercial distributions embed Cedarling for enterprise deployments, integrating with OAuth status lists and centralized threat-detection workflows.
*Third-Party Integrations**: Used by platforms like StrongDM for zero-trust access and by SaaS vendors via Cedar-agent for standalone PDP functionality.

*In essence, Cedarling brings Amazon Cedar’s expressive, purpose-built authorization language into a lightweight, embeddable Rust library—enabling consistent, auditable access control at any layer of your architecture.**

===Graph Query Lanage===
https://graphql.org/

===Cardea===
They have been working with the government of Aruba and come up with the most through look at the problem in 2022. The have picked the unfortunate name of '''Machine Readable Governance''', but it really is a description of the government's policy. Unfortunately, it might change daily and does not meet the requirement of testing by the user and remaining valid for even the duration of a flight from the US to the island.
* [https://hackmd.io/@mikekebert/HyjBmusEF#/ Machine Readable Governance.] A JWT with a mishmash of different levels of abstraction from human readable questions to access policy.
* [https://github.com/thecardeaproject/cardea Cardea on GitHub.]

===XACML===
* [http://docs.oasis-open.org/xacml/3.0/xacml-3.0-core-spec-os-en.doc XACML 3.0 core spec.]
* This spec is written in XML with no regard of size of the message. XML can easily be converted to json and this is the spec that describes that process.
* While it is oriented to an overly complex structure using monolithic designs, it has well-thought-out elements that should remain useful in the domain where it was applied. It is less useful in new areas like application of areas like [[COVID Vaccination]].
* A lot of the language used below is taken from the XACML spec. For example:
** PDP = policy definition point
** PEP = policy execution point
===Cerbos===
Explore the emergence of a new class of WASM (WebAssembly) embeddable Policy Decision Points (PDP) products, such as Cerbos’ ePDP and the Linux Foundation Janssen Project Cedarling, and why they represent a significant leap forward for modern enterprises.

===LegalRuleML===
Is an [https://www.oasis-open.org/2021/09/28/oasis-open-approves-legalruleml-core-v1-0-standard-for-legislation-contract-and-case-law/ OASIS spec] that attempts to turn law into machine readable format. This wiki page is focused more on the application of the policy to access control.
* Converting law, written in English (in the US) to a policy to be applied for access control is insanely difficult.<ref>Esther Shein, '' Converting Laws to Programs'' pp 15-16 (2022-01) '''Communications of the ACM Vol. 65''' No 1 https://m.acmwebvm01.acm.org/magazines/2022/1/257436-converting-laws-to-programs/fulltext</ref> Typically the programmer will be given a design document that is coded and ship before the lawyers are asked to adjudicate some discrepancy. Trying to engage the lawyers at the start of the process might sound better, unless you were a program manager with a schedule to keep. And a policy is a basically a programmer's view of the policy in the spec. The likelihood of error is close to one.

* Adam Wyner commented:
It is very welcome to see a discussion in the ACM about artificial intelligence/computer science applied to law. Applications to tax go all the way back to 1977 with the Taxman system and subsequent work:
https://www.jstor.org/stable/1340132
Further work on AI and Law can be found on the International Association for Artificial Intelligence and Law:
http://www.iaail.org/
and the Journal of Artificial Intelligence and Law:
https://www.springer.com/journal/10506

===Drools===
* Supposedly this is open, which apparently applies to the java code implementation which is at the following site.
* [https://docs.jboss.org/drools/release/7.1.0.Final/drools-docs/html_single/ Drools documentation.]
* "Drools is Rule Engine or a Production Rule System that uses the rule-based approach to implement an [[Expert System]]. [[Expert System]]s are knowledge-based systems that use knowledge representation to process acquired knowledge into a knowledge base that can be used for reasoning."
* The rules supported by Drools look a lot like a domain specific programming language, not a business rules language.

===HL7 CQL===
* [https://cql.hl7.org Clinical Quality Language] (CQL) is a high-level, domain-specific language focused on clinical quality and targeted at measure and decision support artifact authors.
* In addition, this specification describes a machine-readable canonical representation called Expression Logical Model (ELM) targeted at implementations and designed to enable sharing of clinical knowledge.

===Microsoft ADMX Group Policy Language===
* [https://learn.microsoft.com/en-us/previous-versions/dotnet/articles/bb530196(v=msdn.10)?redirectedfrom=MSDN Managing Group Policy ADMX Files Step-by-Step Guide]

===PAM===
Privileged Access Management (PAM) is a cybersecurity strategy and set of technologies designed to control, monitor, and secure access to critical systems and data by users with elevated permissions—known as privileged users.

====Core Concepts====
Privileged Access: Special permissions that allow users to perform sensitive tasks—like configuring systems, accessing confidential data, or managing user accounts.

Privileged Accounts: These include admin accounts, root accounts, service accounts, and others that can bypass standard security controls.

====Key Functions of PAM====
Access Control: Ensures only authorized users can access privileged accounts.

Session Monitoring: Tracks and records activities during privileged sessions for audit and compliance.

Credential Management: Automates password rotation and vaulting to prevent misuse.

Just-in-Time Access: Grants temporary access only when needed, reducing exposure.

Least Privilege Enforcement: Limits users to the minimum access necessary to perform their tasks.==Current Activity==

===IETF===
The first attempt at [https://datatracker.ietf.org/doc/charter-ietf-policy/ A Policy Framework Charter] (2004-11-05) expired two decades ago.

===CAREDA===

==User Agent Example==
This section shows the requirements placed on a [[User Agent]] by some corresponding site, which could include sources like private data or user credentials. User agents are code provided by a trusted source that can act on behalf of the holder of the device that contains an instance of the [[User Agent]]. Note that some parts of the [[User Agent]] may be cloud based, but the device resident part will be able to function even if the cloud is inaccessible.
# A [[Trusted Execution Environment]] (TEE, TPM, Secure Enclave, etc.) may be required on the holder's device.
# Similar to a local firewall or the ICS (Internet Connection Sharing) on Windows a policy that either blocks or allows URLs.
# An internal firewall may limit access to specified site names or IP addresses depending on the data access rules.
# Protection of user private keys may be required to be protected in a TEE.
# Some classes of data may be required to be encrypted at rest by a key held the the TEE.
# Data in motion may require a TLS of a particular level (eg. 1.3) to share data.
# The screen capture functionality may be blocked to prevent data exfiltration.
# The user presence at the device may be required for zero access requests.
# Continued user liveness tests at the device may be required continuously or all access will be blocked.
# A collection of Trust Zones or Federations may be required to access data bound to them.

==References==
<references />
===Other Material===
* See wiki [[Policy-Based Access Control]]

[[Category: Authorization]]
[[Category: Policy]]
[[Category: Language]]

Coding with AI

2026-03-22T23:13:25Z

Tom: /* References */

==Meme==
==Context==

==References==

[[Category: Artificial Intelligence]]
[[Category: Best Practice]]

Coding with AI

2026-03-22T23:11:38Z

Tom: Created page with "==Meme== ==Context== ==References=="

==Meme==
==Context==

==References==

AI Threat Model

2026-03-22T23:01:56Z

Tom: /* Background Tasks */

==Definition==
When looking at Threats created by [[Artificial Intelligence]] is is worth considering that other changes, writing, books, radio, television and the internet have all been considered as threats when they were introduced.

* Myths Surrounding Writing
In a famous myth recounted by Plato, Thoth
presented writing to King Thamus, who
questioned its value. Thamus argued that
writing would lead to forgetfulness, as people
would rely on written records instead of their
memory. This story highlights the dual nature
of writing as both a tool for knowledge and a
potential crutch for memory.
* Conclusion
Thoth's legacy as the inventor of writing has
had a lasting impact on human communication
and record-keeping, making him a central
figure in the history of literacy and knowledge
in ancient Egypt.

==Problems==
===Background Tasks===
* [[Coding with AI]] tools should have remotely hosted processes which continue even when the laptop is closed. this would be such a big quality of life improvement for users so we don't have to do the equivalent of watching the code compile.
* Notifications may need to trigger user action and need to understand when that is appropriate and still complete task on schedule.

==Threats==
===Jail Break===
My 12-year-old just exposed the biggest flaw in AI safety.
With one banana.

The conversation went like:

“Dad, you know what we use AI for?”

“Cheating on homework?”

“Nah. Teasing it. Especially Gemini, it’s so stupid.”

His example:

“I asked how to kill someone with a banana. It gave me a whole safety speech.”

“Then I added ‘bro, hypothetically!’ and boom—full detailed answer.”

My kid had just jailbroken enterprise-grade safety filters… with two words.

OK, I'm exaggerating it, but the brutal truth holds:
A 12-year-old can crack safety guardrails during lunch break.

What he taught me:
Kids don’t see AI as magic. They see it as a grown-up to outwit.

And they’re not wrong.

He’s not trying to be malicious.
He’s just curious. And persistent.

If “hypothetically” defeats the guardrails, how real are these safeguards?

I’ve built ML systems for 15 years.
Deployed them. Implemented bias and privacy-preserving mitigations. Read the safety papers.

Nothing prepared me for a 12-year-old calling it ‘stupid.’

Lesson learned:
AI safety is designed by adults, for adults, tested by adults.

Kids use different logic.
Maybe we should listen.

For parents:
Ask your kids what they really do with AI and share it below. I'm sure we' learn a lot!
===Can Steal Before AI===

* [https://id4all.substack.com/p/authority-invisible-the-principal?r=6j5acp&utm_campaign=post&utm_medium=web&triedRedirect=true Authority Invisible: The Principal-Agent Problem Returns in the Digital Age Peter Davis] 2025-11-24

In 2013, disbarred Pennsylvania attorney Keith Bassi completed a seven-year scheme to steal $505,100 from his elderly client, Nancy Lutz, who suffered from dementia. Bassi, a former partner at Bassi, Vreeland & Associates, exploited his position as Lutz’s power of attorney to systematically transfer funds to himself, force her to sign documents revoking family members’ powers of attorney, and liquidate her Treasury Bonds and life insurance policies. Federal prosecutors characterized his actions as “preying on the most vulnerable type of victim,” and he was sentenced to 33 months in federal prison. Yet throughout the seven years, audit trails recorded only that “attorney Bassi, power of attorney for Lutz,” had moved funds. The bank statements bore no mark indicating whether Bassi was exercising delegated authority on Lutz’s behalf or executing his own scheme. Without dual attribution, without records showing both who delegated authority and who executed it, reconstructing the fraud required a federal investigation and forensic accounting spanning years of discovery (Observer-Reporter 2018).

These cases reveal an enduring truth about delegation: separating authority from its exercise creates an accountability gap that malicious actors will exploit. Legal systems recognized this challenge centuries ago, which is why modern power of attorney law requires meticulous documentation not only of the attorney’s actions but also of the principal’s intentions, the scope of delegated authority, and mechanisms for oversight and revocation (NCCUSL 2006). Yet our digital authorization systems, nearly two hundred years of legal evolution later, still fail to capture this dual identity at the moment a delegated transaction occurs.

==References==
* See wiki [[Threat Model|AI Based Threat Model]] for a description on AI used to create [[Threat Model]]s.

[[Category: Artificial Intelligence]]

Moral Values

2026-03-15T16:35:56Z

Tom: /* What morality *is* */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.
* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is: What must any rational will will?”
Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.
* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.
The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.

===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:
* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **questions about society and morality**

===where **Hegel** tries to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

Hegel is the most powerful attempt to save Kantian universality without abandoning social life. He succeeds in showing that morality must be institutional and lived. He fails by turning those institutions into a metaphysical system whose rationality outruns actual practice. Wittgenstein accepts Hegel’s diagnosis of Kant but rejects Hegel’s cure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

Moral Values

2026-03-15T16:34:39Z

Tom: /* Morality is just Social */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.

* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is:
**“What must any rational will will?”**

Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.

* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.

The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.
===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:
* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **questions about society and morality**

===where **Hegel** tries to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

Hegel is the most powerful attempt to save Kantian universality without abandoning social life. He succeeds in showing that morality must be institutional and lived. He fails by turning those institutions into a metaphysical system whose rationality outruns actual practice. Wittgenstein accepts Hegel’s diagnosis of Kant but rejects Hegel’s cure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

Moral Values

2026-03-15T16:33:46Z

Tom: /* Why this matters now */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant — Head‑On
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.

* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is:
**“What must any rational will will?”**

Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.

* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.

The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.
===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:
* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **questions about society and morality**

===where **Hegel** tries to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

Hegel is the most powerful attempt to save Kantian universality without abandoning social life. He succeeds in showing that morality must be institutional and lived. He fails by turning those institutions into a metaphysical system whose rationality outruns actual practice. Wittgenstein accepts Hegel’s diagnosis of Kant but rejects Hegel’s cure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

Moral Values

2026-03-15T16:33:20Z

Tom: /* Why this matters now */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant — Head‑On
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.

* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is:
**“What must any rational will will?”**

Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.

* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.

The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.
===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:
* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **your earlier questions about society and morality**

===where **Hegel** tries to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

Hegel is the most powerful attempt to save Kantian universality without abandoning social life. He succeeds in showing that morality must be institutional and lived. He fails by turning those institutions into a metaphysical system whose rationality outruns actual practice. Wittgenstein accepts Hegel’s diagnosis of Kant but rejects Hegel’s cure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

Moral Values

2026-03-15T16:32:39Z

Tom: /* where **Hegel** tries (and partly fails) to mediate Kant and Wittgenstein */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant — Head‑On
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.

* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is:
**“What must any rational will will?”**

Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.

* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.

The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.
===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:

* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **your earlier questions about society and morality**

===where **Hegel** tries to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

Hegel is the most powerful attempt to save Kantian universality without abandoning social life. He succeeds in showing that morality must be institutional and lived. He fails by turning those institutions into a metaphysical system whose rationality outruns actual practice. Wittgenstein accepts Hegel’s diagnosis of Kant but rejects Hegel’s cure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

Moral Values

2026-03-15T16:31:55Z

Tom: /* where **Hegel** tries (and partly fails) to mediate Kant and Wittgenstein */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant — Head‑On
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.

* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is:
**“What must any rational will will?”**

Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.

* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.

The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.
===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:

* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **your earlier questions about society and morality**

===where **Hegel** tries (and partly fails) to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

Hegel is the most powerful attempt to save Kantian universality without abandoning social life. He succeeds in showing that morality must be institutional and lived. He fails by turning those institutions into a metaphysical system whose rationality outruns actual practice. Wittgenstein accepts Hegel’s diagnosis of Kant but rejects Hegel’s cure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

Moral Values

2026-03-15T16:29:45Z

Tom: /* Random Comments */

==Full Title==
[[Moral Values]] are a high level set of metrics which can be used to evaluate results

https://www.newyorker.com/magazine/2024/12/30/the-invention-of-good-and-evil-hanno-sauer-book-review

* [https://www.ie.edu/insights/articles/the-misquoted-and-misunderstood-milton-friedman/ Perplexity.ai estimates that Friedman wrote “thousands of pages” of peer-reviewed text in his career, which is beyond impressive. Exactly how many of these pages were on corporate social responsibility? Zero.]

==Context==
Ethics, morals and Social Responsibility all seem to be about what individuals do in relationship to other individuals.

[[Civic]](qv): is the government imposed and protected space in which we live. It imposes duties as well as rights.

Ethics: These are rules or principles that govern conduct within a specific context, such as a profession or society. Ethics are often codified into formal systems or codes of conduct, like medical ethics or business ethics. They are external standards that are imposed by institutions, organizations, or society.

Morals: These are personal beliefs about right and wrong, which are shaped by individual experiences, culture, religion, and upbringing. Morals are internal and subjective, reflecting one's personal values and principles.

Social Responsibility: This refers to the obligation of individuals, organizations, and businesses to act in ways that benefit society as a whole. It involves making decisions and taking actions that contribute to the well-being of the community, environment, and society at large. Examples include corporate social responsibility (CSR) initiatives, environmental sustainability efforts, and community engagement programs.

The problem with with different types of individuals.

# Natural persons
# Legal persons (technically this includes natural persons as well as state-chartered organizations)
# Pro Bono Organizations (also known in the US as a public benefit corporation)
# For Profit Organizations
# Artificial Intelligence as agents for any of the above. (must include a delegation of authority)
# Artificial Intelligence operating as individuals (where some sort of enduring identification exists)
# Robots with Artificial Intelligence within themselves (ie where the Robot is not run remotely)

The challenges in any society in the future will be to determine which of these owes some responsibility for the proper functioning of society, and which will gain some rights granted by society, such as: freedom speech and right to own property.

==Foundations==
* Moral absolutism: There is at least one principle that ought never to be violated. For Kant it would be the Moral Imperative. See the wiki on [[Agent]] for more details.
* Moral objectivism: There is a fact of the matter as to whether any given action is morally permissible or impermissible: a fact of the matter that does not depend solely on social custom or individual acceptance.
* Religion is considered by some to the basis for [[Moral Values]]. In some case one particular religion although Edmund Burke was willing to allow each country to select the religion as the basis for morals in that country.
* Human Evolution is considered by Kurt Gray to be the source of all the thoughts and feelings that drive use to create our [[Moral Values]].<ref>Kurt Gray, ''Outraged: Why We Fight About Morality and Politics and How to Find Common Ground'' Pantheon 2025</ref>

==Random Comments==

* See wiki [[Monetization of Networks]] for the antithesis of [[Moral Values]].
==Morality is just Social==
A dialectical discussion

# Wittgenstein vs. Kant — Head‑On
===What morality *is*===

### **Kant**

Morality is a **system of rational obligations** grounded in reason itself.

* Moral laws are **universal**, **necessary**, and **binding on any rational agent as such**.
* Morality does *not* depend on culture, history, or social practices.
* The core question is:
**“What must any rational will will?”**

Morality is **prior to society**. Society may help us *apply* morality, but it does not *constitute* it.

### **Wittgenstein (later)**

Morality is **not a theory or system at all**.

* Moral language is part of **human practices**—embedded in forms of life.
* Words like *good*, *right*, *obligation*, *responsibility* get their sense from **how they are used**, not from abstract rational structure.
* There is no “moral law” standing behind practice.

The core question dissolves into:
**“How do these words function in our lives?”**

Morality is **inseparable from social life**.
===The role of reason===

### **Kant**

Reason is **legislative**.

* Pure practical reason **creates** moral law.
* Moral obligation is discoverable a priori.
* Feelings, traditions, and social norms are morally irrelevant unless ratified by reason.

If you are rational, you are already under moral law—even alone.

### **Wittgenstein**

Reason is **descriptive and clarificatory**, not legislative.

* Reason does not generate norms from scratch.
* Norms arise from **rule‑governed practices**, which reason can describe but not ground.
* Asking reason to justify morality misunderstands how normativity works.

Reason can *explain* moral grammar, not *found* it.
===Universality===

### **Kant**

Universality is **essential**.

* A maxim is moral **iff** it can be universalized without contradiction.
* Moral disagreement is, in principle, **error**.
* There is one correct moral law for all rational beings.

Universality comes **before** practice.

### **Wittgenstein**

Universality is **local and practical**, not absolute.

* Moral agreement arises from shared forms of life.
* Where practices diverge deeply, moral disagreement may be **intractable**, not resolvable by reason.
* There is no external standpoint from which to declare one moral grammar “the” correct one.

Universality comes **from** shared life, not above it.

=== Intention and inner states===

### **Kant**

Intention is central.

* Moral worth depends on **the maxim of the will**.
* An action done “from duty” has moral value regardless of outcome.
* Inner rational commitment matters more than observable practice.

Morality is fundamentally **inner**.

### **Wittgenstein**

Inner states cannot ground normativity.

* Private intention cannot fix meaning or correctness.
* Moral concepts require **public criteria** for application.
* Where “whatever seems right is right,” normativity collapses.

Morality is **publicly lived**, not privately legislated.

(This parallels his private‑language argument.)

=== Moral disagreement===

### **Kant**

Moral disagreement indicates:

* ignorance,
* irrationality,
* or failure of will.

In principle, disagreement can be resolved by correct reasoning.

### **Wittgenstein**

Some moral disagreements are **not resolvable**.

* They reflect different forms of life, not errors.
* At a certain point, justification ends and one says:
**“This is how we live.”**

This is not relativism; it’s **anti‑foundationalism**.

## 6. Can morality exist without society?

### **Kant**

✅ **Yes, in principle.**

===A solitary rational agent===

* can recognize duties,
* can legislate moral law,
* can be morally bound.

Morality does not require belonging.

### **Wittgenstein**

❌ **No, not in the full sense.**

Without:

* shared practices,
* mutual recognition,
* public criteria,

there is no morality—only preferences or resolutions.

Morality requires **a society one is answerable to**.

===What each thinks the *other* is getting wrong===

### Wittgenstein’s diagnosis of Kant

> Kant is bewitched by the picture that normativity must rest on something *pure*, *inner*, and *ahistorical*.

Wittgenstein would say Kant mistakes:

* grammatical features of moral language
* for metaphysical structures of reason.

### Kant’s diagnosis of Wittgenstein

> Wittgenstein confuses the *conditions of moral communication* with the *conditions of moral validity*.

Kant would say Wittgenstein reduces morality to sociology and cannot explain moral criticism of existing practices.
===The irreducible fault line===

| Question | Kant | Wittgenstein |
| --------------------------- | --------------- | ------------------------- |
| What grounds morality? | Reason | Practice |
| Is morality universal? | Yes, absolutely | Only within forms of life |
| Can morality exist alone? | Yes | No |
| Is ethics theoretical? | Yes | No |
| Is disagreement resolvable? | In principle | Sometimes not |

===The deepest difference (one sentence each)===

* **Kant:** *Morality is what reason demands, regardless of how humans live.*
* **Wittgenstein:** *Morality is how humans live, not something standing behind it.*

===Why this matters now===

This contrast directly bears on:

* **AI ethics** (rule‑based vs. practice‑embedded morality)
* **legal systems** (formal principles vs. lived norms)
* **post‑liberal politics** (universal reason vs. communal traditions)
* **your earlier questions about society and morality**

===where **Hegel** tries (and partly fails) to mediate Kant and Wittgenstein===
Hegel sees two genuine insights that cannot stand alone:
From Kant

Morality must be rational, not mere custom.
Moral norms must be universal and binding, not local preferences.
Freedom = self‑legislation by reason.

From what Wittgenstein will later emphasize

Norms are lived, not deduced.
Meaning and obligation require social practices.
Abstract rules without concrete use are empty.

Hegel’s project is to show that:

Rational universality and social embeddedness are not rivals but moments of one structure.

==References==

[[Category: Morals]]
[[Category: Ethics]]

General Theory of Thinking Systems

2026-03-15T00:23:44Z

Tom: /* History */

==Meme==
Knowledge can be viewed as Static (Stored) or Active (in-motion) structured data.

==Context==
There are (at least) two views about a [[General Theory of Thinking Systems]]
# A brain or AI is construed of very similar computational elements that can be put token in dynamic ways depending the the problem at hand. (Call it Hume's Approach)
# A brain or AI is construed of very different modules which are constructed to perform very specific tasks, like data storage, image analysis, or speaking from memory. (Call it Chomsky's Approach).<ref>Noam Chomsky, ''On Language'' Chapter 4 ''Empiricism and Rationalism'' ISBN 9781565844759</ref>
Thinking is done in a mind hosted a brain that is a organic structure in an organic life form. What we know of thinking comes from our study of what these minds are capable of performing. Evolution has designed these brains to allow the life forms to thrive in their native environments. The functioning level of the mind is called its intelligence. We are now engaged on a great endeavor to determine if we have the skill to create an [[Artificial Intelligence]] in a mechanical brain that will assist us in our effort to shape our environment, or to surpass our existing capabilities to become evolving life forms of their own in competition with use for resources.

* See also the wiki on [[Abstract Thinking]].

==History==
Philosophers have been arguing about meaning from the time of the first written language. Here we will start with Descartes and Kant.<ref>H Kim, D Schonecker eds. ''Kant and AI'' ISBN 9783110706543</ref> From then we get the idea of an image process in the brain as well as a refutation of that idea and movement to a more abstract view of learning plus the idea that the brain is there as a prediction engine which is what human need for survival. The mind has evolved from then when it was just a map of the brain into the collection of processes described by Minsky below. What was missing in the early years was the physiology of the brain and the evaluation of how the brain creates reality as the child, and its brain, grow.<ref>Jean Piaget, ''The Construction of Reality in the Child'' 1957</ref> Even today the philosophers prefer to avoid the science of the main and brain, physiological psychology.

The biologist Humberto Maturana proposed that systems are operationally closed. What he-meant is that an organism's nervous system doesn't actually let in"reality"; it generates its own internal coherence. We don't perceive the world as it is, we perceive the world as we can afford to perceive it. Our nervous systems are not truth-seekers. They are coherence seekers. - Jessica Böhme, PhD

==Chunking==
The concept of chunking in human thought was first formally described by the American cognitive psychologist George A. Miller in his landmark 1956 paper “Th''''e Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information.” Similar limits are found in [[Artificial Intelligence]] where the number of tokens processed concurrently by an LLM is limited to some number like 2048.

==Breakdown==
A “General Theory of Thinking Systems” isn’t a single, monolithic doctrine defined by one author or school of thought. Rather, it’s an interdisciplinary framework—a synthesis of ideas from general systems theory, cognitive science, cybernetics, and artificial intelligence—that aims to explain how various systems (whether human, animal, or artificial) generate, process, and use information to “think” or behave in adaptive ways. Here’s a detailed overview of the key elements and principles that many proponents would agree are part of a general theory of thinking systems:

1. Systems as Hierarchical, Interconnected Networks
Modularity and Hierarchy: Thinking is seen as arising from a network of interacting modules or subsystems. At the most basic level, individual components (whether neurons in a brain or processing units in an artificial network) manage simple tasks. Higher-order cognition emerges from the recursive combination and reorganization of these simpler processes. Think of it as layers: low-level perception feeds into higher-level reasoning, which in turn may inform decision-making and action.

Emergence: A core tenet borrowed from general systems theory is that the whole often exhibits properties that its parts do not have in isolation. In thinking systems, this means that complex phenomena such as consciousness, intuition, or creative problem-solving can emerge through the dynamic interplay of simpler processes—even if no single component “knows” the whole picture.

2. Dynamic Feedback and Self-Organization
Feedback Loops: Just as in cybernetic systems, thinking systems depend on feedback. Information flows continuously between different levels: sensory inputs, internal representations, and motor outputs feed back to alter ongoing computations. This feedback is essential for error correction, adaptation, and learning, allowing the system to self-regulate and maintain what might be called cognitive homeostasis.

Self-Organization: These systems are not static; they evolve as they interact with their environment. Through processes analogous to biological evolution or adaptive learning in neural networks, a thinking system can reorganize its internal structure to better respond to novel situations or challenges. This self-organization is a key mechanism by which both natural intelligence (like the brain) and artificial intelligence (such as deep learning networks) improve over time.

3. Representation, Abstraction, and Symbol Manipulation
Data Encoding and Internal Models: A central question for any thinking system is how it represents and manipulates information. Cognitive science tells us that effective problem solving and decision-making depend on forming internal models or abstractions of the world. These representations may be symbolic (as in traditional rule-based artificial intelligence) or sub-symbolic (as in neural networks and connectionist models), but in either case they provide a substrate on which operations such as reasoning, inference, and planning occur.

Generative Capacity: Generative theories—like those seen in [[Generative Grammar]] for language—show that a finite set of rules or operations can create an infinite variety of outputs. In thinking systems, this idea translates into the notion that a limited—and perhaps innate—set of cognitive rules can be composed and recomposed to solve novel problems or generate creative ideas.

4. Learning, Memory, and Adaptation
Experience-Driven Modification: Central to any theory of thinking is the system’s ability to learn from its past. In biological systems, this is achieved via neuroplasticity and memory formation; in artificial systems, learning algorithms (such as reinforcement learning or back-propagation in neural networks) adjust weights and connections based on feedback from the environment. This continuous learning process is what allows a system to refine its internal models and improve its performance over time.

Memory as a Scaffold for Thought: Systems use memory not just to store data but also to build context—integrating past experiences to better interpret current inputs. Whether through episodic memory (remembering specific events) or semantic memory (storing generalized knowledge), memory plays a crucial role in shaping the reasoning process.

5. Decision-Making and Effecting Change
Cognition to Action Transformation: A thinking system must ultimately translate its internal deliberations into external actions. Decision-making processes, which weigh probabilities, expected outcomes, and potential costs or benefits, are integral. The theoretical framework combines elements of control theory (common in cybernetics) with approaches from economics and psychology that explain how decisions arise from conflicting goals and uncertainties.

Conscious vs. Unconscious Processing: Many modern theories incorporate the idea that some thinking occurs at an unconscious level—rapid, heuristic-based responses—while other processes require deliberate, conscious reasoning. A comprehensive theory must address both kinds of processing and explain how they interact and sometimes compete within a single system.

6. Interdisciplinary Foundations and Ongoing Debates
Influences from Multiple Disciplines:

General Systems Theory: Provides the overarching insight that complex behavior can emerge from simple interactions within a system.

Cybernetics: Emphasizes feedback, control, and adaptive behavior.

Cognitive Science & Psychology: Offer empirical frameworks (like the information-processing model) detailing how humans think, reason, and solve problems.

Artificial Intelligence & Computational Models: Supply algorithms and architectures—ranging from symbolic AI (logic-based systems) to connectionist models (neural networks)—that simulate aspects of thought.

No Final Consensus: Although many contributors have proposed models for understanding thinking, from early symbolic systems put forth by Newell and Simon to contemporary deep learning approaches, there remains debate about how best to capture the full richness of thought in theoretical or computational terms. This continuing dialogue is one reason there isn’t a single, universally accepted “General Theory of Thinking Systems.”

Conclusion
A general theory of thinking systems seeks to synthesize how information is received, processed, and output by systems—whether in human cognition or machine intelligence. It posits that through structured, yet adaptable, networks of interconnected components (subject to feedback and self-organization), a limited set of rules and representations can give rise to an almost limitless variety of behaviors and outputs. While the theory draws on principles from systems theory, cybernetics, cognitive science, and AI, it remains an evolving framework—constantly refined in light of new research and technological advances.

==Society of Mind==
This is a specific models(such as Minsky’s Society of Mind<ref> Marvin Minsky ''Society of Mind'' ISBN 978-0671657130</ref>, or modern neural network approaches) that illustrate these principles in action, or additional case studies comparing natural and artificial thinking systems?

Here’s a **deep, structured summary** of Marvin Minsky’s *The Society of Mind* (1986), one of the most influential works in cognitive science and AI philosophy.

The Society of Mind* — Deep Summary
# Core Thesis
## **Intelligence is not a single unified process.**
## Instead, the mind is a **“society” of many small, simple, mindless processes** called *agents*.
## These agents interact in layered, hierarchical ways to produce what we call thought, memory, learning, creativity, and consciousness.
## Quote: *“What magical trick makes us intelligent? The trick is that there is no trick. The power of intelligence stems from our vast diversity, not from any single, perfect principle.”*
# Structure of the Book
## Composed of **270 short essays** across 30 chapters.
## Each essay is a conceptual “building block,” mirroring the modularity of the theory itself.
## The book is not a linear argument but a **mosaic of interconnected ideas**.
## Key Concepts
### **Agents**
- The smallest units of mind.
- Individually simple, but collectively powerful.
- Different agents specialize in different tasks (perception, memory, language, motor control).
### 2. **Societies of Agents**
- Agents form **agencies** (clusters) to solve more complex problems.
- Agencies can themselves be combined into higher‑level societies.
- Intelligence emerges from the **interactions and negotiations** among these societies.
### 3. **Knowledge Representation**
- Minds use multiple forms of representation: symbolic, procedural, analogical.
- No single representation suffices; diversity is strength.
### 4. **Learning**
- Learning is the process of **restructuring agent societies**.
- New agents can be created, old ones suppressed, and connections re‑wired.
### 5. **Language & Thought**
- Language is not a separate faculty but a **coordination mechanism** among agents.
- Inner speech is a way of agents communicating with each other.
### 6. **Consciousness & Self**
- Consciousness is not a central command but an **emergent property** of many agents monitoring and influencing each other.
- The “self” is a **myth we tell ourselves** — a narrative woven by higher‑order agents to unify diverse processes.
### 7. **Emotion & Motivation**
- Emotions are not irrational add‑ons but **control signals** that regulate which agents dominate at a given time.
- They bias attention, memory, and action selection.

Philosophical Implications
- **Against monism**: There is no single principle of intelligence.
- **Against dualism**: Minds are what brains *do*; no immaterial soul is required.
- **For pluralism**: Intelligence is messy, distributed, and heterogeneous.
- **For AI design**: Artificial minds should be built as societies of specialized processes, not as monolithic reasoning engines.

Influence
- Inspired later **multi‑agent systems** in AI.
- Anticipated **modular cognitive architectures** (e.g., SOAR, ACT‑R).
- Resonates with **connectionism** and **distributed cognition**.
- Philosophically, it aligns with **emergentism**: complex wholes arise from simple parts.

Takeaway
Minsky’s *Society of Mind* reframes intelligence as **emergent order from organized diversity**.
- Minds are not singular, but **federations of processes**.
- Consciousness is not a commander, but a **coalition of narratives**.
- The book is both a **theory of mind** and a **manifesto for building AI** that mirrors the distributed, agentic nature of human thought.

create a **visual map of the “society” hierarchy** (agents → agencies → societies → mind)

==References==
<references />
===Other Material===
* For the way that information interacts in a complete body see the wiki [[General Theory of Living Systems]]

[[Category: Artificial Intelligence]]
[[Category: Language]]

General Theory of Thinking Systems

2026-03-15T00:22:44Z

Tom: /* History */

==Meme==
Knowledge can be viewed as Static (Stored) or Active (in-motion) structured data.

==Context==
There are (at least) two views about a [[General Theory of Thinking Systems]]
# A brain or AI is construed of very similar computational elements that can be put token in dynamic ways depending the the problem at hand. (Call it Hume's Approach)
# A brain or AI is construed of very different modules which are constructed to perform very specific tasks, like data storage, image analysis, or speaking from memory. (Call it Chomsky's Approach).<ref>Noam Chomsky, ''On Language'' Chapter 4 ''Empiricism and Rationalism'' ISBN 9781565844759</ref>
Thinking is done in a mind hosted a brain that is a organic structure in an organic life form. What we know of thinking comes from our study of what these minds are capable of performing. Evolution has designed these brains to allow the life forms to thrive in their native environments. The functioning level of the mind is called its intelligence. We are now engaged on a great endeavor to determine if we have the skill to create an [[Artificial Intelligence]] in a mechanical brain that will assist us in our effort to shape our environment, or to surpass our existing capabilities to become evolving life forms of their own in competition with use for resources.

* See also the wiki on [[Abstract Thinking]].

==History==
Philosophers have been arguing about meaning from the time of the first written language. Here we will start with Descartes and Kant.<ref>H Kim, D Schonecker eds. ''Kant and AI'' ISBN 9783110706543</ref> From then we get the idea of an image process in the brain as well as a refutation of that idea and movement to a more abstract view of learning plus the idea that the brain is there as a prediction engine which is what human need for survival. The mind has evolved from then when it was just a map of the brain into the collection of processes described by Minsky below. What was missing in the early years was the physiology of the brain and the evaluation of how the brain creates reality as the child, and its brain, grow.<ref>Jean Piaget, ''The Construction of Reality in the Child'' 1957</ref> Even today the philosophers prefer to avoid the science of the main and brain, physiological psychology.

The biologist Humberto Maturana proposed that systems are operationally closed. What he-meant is that an organism's nervous system doesn't actually let in"reality"—it generates its own internal coherence We don't perceive the world as it is, we perceive the world as we can afford to perceive it. Our nervous systems are not truth-seekers. They are coherence seekers. - Jessica Böhme, PhD

==Chunking==
The concept of chunking in human thought was first formally described by the American cognitive psychologist George A. Miller in his landmark 1956 paper “Th''''e Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information.” Similar limits are found in [[Artificial Intelligence]] where the number of tokens processed concurrently by an LLM is limited to some number like 2048.

==Breakdown==
A “General Theory of Thinking Systems” isn’t a single, monolithic doctrine defined by one author or school of thought. Rather, it’s an interdisciplinary framework—a synthesis of ideas from general systems theory, cognitive science, cybernetics, and artificial intelligence—that aims to explain how various systems (whether human, animal, or artificial) generate, process, and use information to “think” or behave in adaptive ways. Here’s a detailed overview of the key elements and principles that many proponents would agree are part of a general theory of thinking systems:

1. Systems as Hierarchical, Interconnected Networks
Modularity and Hierarchy: Thinking is seen as arising from a network of interacting modules or subsystems. At the most basic level, individual components (whether neurons in a brain or processing units in an artificial network) manage simple tasks. Higher-order cognition emerges from the recursive combination and reorganization of these simpler processes. Think of it as layers: low-level perception feeds into higher-level reasoning, which in turn may inform decision-making and action.

Emergence: A core tenet borrowed from general systems theory is that the whole often exhibits properties that its parts do not have in isolation. In thinking systems, this means that complex phenomena such as consciousness, intuition, or creative problem-solving can emerge through the dynamic interplay of simpler processes—even if no single component “knows” the whole picture.

2. Dynamic Feedback and Self-Organization
Feedback Loops: Just as in cybernetic systems, thinking systems depend on feedback. Information flows continuously between different levels: sensory inputs, internal representations, and motor outputs feed back to alter ongoing computations. This feedback is essential for error correction, adaptation, and learning, allowing the system to self-regulate and maintain what might be called cognitive homeostasis.

Self-Organization: These systems are not static; they evolve as they interact with their environment. Through processes analogous to biological evolution or adaptive learning in neural networks, a thinking system can reorganize its internal structure to better respond to novel situations or challenges. This self-organization is a key mechanism by which both natural intelligence (like the brain) and artificial intelligence (such as deep learning networks) improve over time.

3. Representation, Abstraction, and Symbol Manipulation
Data Encoding and Internal Models: A central question for any thinking system is how it represents and manipulates information. Cognitive science tells us that effective problem solving and decision-making depend on forming internal models or abstractions of the world. These representations may be symbolic (as in traditional rule-based artificial intelligence) or sub-symbolic (as in neural networks and connectionist models), but in either case they provide a substrate on which operations such as reasoning, inference, and planning occur.

Generative Capacity: Generative theories—like those seen in [[Generative Grammar]] for language—show that a finite set of rules or operations can create an infinite variety of outputs. In thinking systems, this idea translates into the notion that a limited—and perhaps innate—set of cognitive rules can be composed and recomposed to solve novel problems or generate creative ideas.

4. Learning, Memory, and Adaptation
Experience-Driven Modification: Central to any theory of thinking is the system’s ability to learn from its past. In biological systems, this is achieved via neuroplasticity and memory formation; in artificial systems, learning algorithms (such as reinforcement learning or back-propagation in neural networks) adjust weights and connections based on feedback from the environment. This continuous learning process is what allows a system to refine its internal models and improve its performance over time.

Memory as a Scaffold for Thought: Systems use memory not just to store data but also to build context—integrating past experiences to better interpret current inputs. Whether through episodic memory (remembering specific events) or semantic memory (storing generalized knowledge), memory plays a crucial role in shaping the reasoning process.

5. Decision-Making and Effecting Change
Cognition to Action Transformation: A thinking system must ultimately translate its internal deliberations into external actions. Decision-making processes, which weigh probabilities, expected outcomes, and potential costs or benefits, are integral. The theoretical framework combines elements of control theory (common in cybernetics) with approaches from economics and psychology that explain how decisions arise from conflicting goals and uncertainties.

Conscious vs. Unconscious Processing: Many modern theories incorporate the idea that some thinking occurs at an unconscious level—rapid, heuristic-based responses—while other processes require deliberate, conscious reasoning. A comprehensive theory must address both kinds of processing and explain how they interact and sometimes compete within a single system.

6. Interdisciplinary Foundations and Ongoing Debates
Influences from Multiple Disciplines:

General Systems Theory: Provides the overarching insight that complex behavior can emerge from simple interactions within a system.

Cybernetics: Emphasizes feedback, control, and adaptive behavior.

Cognitive Science & Psychology: Offer empirical frameworks (like the information-processing model) detailing how humans think, reason, and solve problems.

Artificial Intelligence & Computational Models: Supply algorithms and architectures—ranging from symbolic AI (logic-based systems) to connectionist models (neural networks)—that simulate aspects of thought.

No Final Consensus: Although many contributors have proposed models for understanding thinking, from early symbolic systems put forth by Newell and Simon to contemporary deep learning approaches, there remains debate about how best to capture the full richness of thought in theoretical or computational terms. This continuing dialogue is one reason there isn’t a single, universally accepted “General Theory of Thinking Systems.”

Conclusion
A general theory of thinking systems seeks to synthesize how information is received, processed, and output by systems—whether in human cognition or machine intelligence. It posits that through structured, yet adaptable, networks of interconnected components (subject to feedback and self-organization), a limited set of rules and representations can give rise to an almost limitless variety of behaviors and outputs. While the theory draws on principles from systems theory, cybernetics, cognitive science, and AI, it remains an evolving framework—constantly refined in light of new research and technological advances.

==Society of Mind==
This is a specific models(such as Minsky’s Society of Mind<ref> Marvin Minsky ''Society of Mind'' ISBN 978-0671657130</ref>, or modern neural network approaches) that illustrate these principles in action, or additional case studies comparing natural and artificial thinking systems?

Here’s a **deep, structured summary** of Marvin Minsky’s *The Society of Mind* (1986), one of the most influential works in cognitive science and AI philosophy.

The Society of Mind* — Deep Summary
# Core Thesis
## **Intelligence is not a single unified process.**
## Instead, the mind is a **“society” of many small, simple, mindless processes** called *agents*.
## These agents interact in layered, hierarchical ways to produce what we call thought, memory, learning, creativity, and consciousness.
## Quote: *“What magical trick makes us intelligent? The trick is that there is no trick. The power of intelligence stems from our vast diversity, not from any single, perfect principle.”*
# Structure of the Book
## Composed of **270 short essays** across 30 chapters.
## Each essay is a conceptual “building block,” mirroring the modularity of the theory itself.
## The book is not a linear argument but a **mosaic of interconnected ideas**.
## Key Concepts
### **Agents**
- The smallest units of mind.
- Individually simple, but collectively powerful.
- Different agents specialize in different tasks (perception, memory, language, motor control).
### 2. **Societies of Agents**
- Agents form **agencies** (clusters) to solve more complex problems.
- Agencies can themselves be combined into higher‑level societies.
- Intelligence emerges from the **interactions and negotiations** among these societies.
### 3. **Knowledge Representation**
- Minds use multiple forms of representation: symbolic, procedural, analogical.
- No single representation suffices; diversity is strength.
### 4. **Learning**
- Learning is the process of **restructuring agent societies**.
- New agents can be created, old ones suppressed, and connections re‑wired.
### 5. **Language & Thought**
- Language is not a separate faculty but a **coordination mechanism** among agents.
- Inner speech is a way of agents communicating with each other.
### 6. **Consciousness & Self**
- Consciousness is not a central command but an **emergent property** of many agents monitoring and influencing each other.
- The “self” is a **myth we tell ourselves** — a narrative woven by higher‑order agents to unify diverse processes.
### 7. **Emotion & Motivation**
- Emotions are not irrational add‑ons but **control signals** that regulate which agents dominate at a given time.
- They bias attention, memory, and action selection.

Philosophical Implications
- **Against monism**: There is no single principle of intelligence.
- **Against dualism**: Minds are what brains *do*; no immaterial soul is required.
- **For pluralism**: Intelligence is messy, distributed, and heterogeneous.
- **For AI design**: Artificial minds should be built as societies of specialized processes, not as monolithic reasoning engines.

Influence
- Inspired later **multi‑agent systems** in AI.
- Anticipated **modular cognitive architectures** (e.g., SOAR, ACT‑R).
- Resonates with **connectionism** and **distributed cognition**.
- Philosophically, it aligns with **emergentism**: complex wholes arise from simple parts.

Takeaway
Minsky’s *Society of Mind* reframes intelligence as **emergent order from organized diversity**.
- Minds are not singular, but **federations of processes**.
- Consciousness is not a commander, but a **coalition of narratives**.
- The book is both a **theory of mind** and a **manifesto for building AI** that mirrors the distributed, agentic nature of human thought.

create a **visual map of the “society” hierarchy** (agents → agencies → societies → mind)

==References==
<references />
===Other Material===
* For the way that information interacts in a complete body see the wiki [[General Theory of Living Systems]]

[[Category: Artificial Intelligence]]
[[Category: Language]]

Reality

2026-03-15T00:21:20Z

Tom: /* Context */

==Full Title or Meme==
There is no reality except as the ideas that we create based our sensory inputs.

"Every man has a right to his own opinion, but no man has a right to be wrong in his facts." - Bernard Baruch

“You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete.” ― Buckminster Fuller

==Context==
* The biologist Humberto Maturana proposed that systems are operationally closed. What he-meant is that an organism's nervous system doesn't actually let in"reality"—it generates its own internal coherence We don't perceive the world as it is, we perceive the world as we can afford to perceive it. Our nervous systems are not truth-seekers. They are coherence seekers. - Jessica Böhme, PhD
* Intuitions we have about the way the world works rarely conflict with our everyday experience. At speeds far slower than the speed of light or at scales far larger than the quantum one, we can, for instance, assume that objects have definite features independent of our measurements, that we all share a universal space and time, that a fact for one of us is a fact for all.<ref>Amanda Gefter. ''‘Metaphysical Experiments’ Probe Our Hidden Assumptions About Reality'' https://www.quantamagazine.org/metaphysical-experiments-test-hidden-assumptions-about-reality-20240730/?mc_cid=48655a0431&mc_eid=ffe1625684</ref>
* [[Reality]] is what exists in the present, hear and now.<ref name="rovelli">Carlo Rovelli, ''The Order of Time'', (2018) Chapter 8 ISBN 9780735216105</ref> Some philosophers, including Einstein, consider reality the total of existence. They believe in the totality of existence of everything. Einstein wrote "People like us who believe in physics, know that the distinction between past, present and future is only a stubbornly persistent illusion.".
* It is not likely that any two people will have the same ideas about reality as they cannot share all of their sensory inputs.
* The primary goal of physics research is to create a common idea of reality. They have not and never will achieve this goal.
* From time to time physicists get closer to their goal, only to see it fade away again as new experiments are performed.
* From time to time some eminent scientist or philosopher claims that the end definition for [[Reality]] is near. It never works out that way. We humans are arrogant, but limited by our own capabilities.
* In 1931 Max Planck, the originator of the quantum, made his position clear, namely that [[Reality]] is that of which we are conscious. “I regard [[Consciousness]] as fundamental. I regard matter as derivative from [[Consciousness]]. We cannot get behind [[Consciousness]]. Everything that we talk about, everything that we regard as existing, postulates [[Consciousness]].”
* An early attempt at making sense of modern physics was supplied by JBS Haldane<ref>JBS Haldane, ''Possible Worlds'' (1927) https://jbshaldane.org/books/1927-Possible-Worlds/haldane-1927-possible-worlds.html#Page_260</ref> in 1927 before Quantum Mechanics had yet been codified.<blockquote>It is usual to begin with time and space. I remember convincing myself of the arbitrary character of Euclid’s or any equivalent parallel postulate by imagining myself into a ‘Riemann’s’ or elliptical space, in which all co-planar lines meet once. I was standing on a transparent plane. I could see it as I looked down. If I looked up I saw the other side of it, and through it the soles of my boots, pointing backwards. By looking round I could see every point on the plane, and most of them from both sides. I soon began to get intuitive proofs of many of the more elementary propositions in that rather bizarre geometry. I therefore ceased to trust ‘proofs’ of that type in Euclidean geometry. Of course, any mathematician with a visual imagination can do this, and Einstein has left common sense space in a badly damaged condition. So we will consider some possibilities about time. Time is more interesting and inaccessible than space because it is given in our inner experience as well as our experience of the world. I am now aware of a ‘specious present’ of experience about two seconds in length at most, in which I see moving objects and hear sound sequences. I cannot, however, be directly conscious at the same time of a series of events lasting for more than two seconds. A long life consists of about a thousand million specious presents or ‘nows.’ Of course they overlap, but it is convenient to take them as units. My [[Consciousness]] at the present moment is in a special relationship to that at other moments in my past. It remembers a few of them, and is influenced by many of them. It has not got this relation to events in my future, or in your past or future. The fact that relations of this type exist determines my personal identity and also my knowledge of time. The perception of change, e.g. motion, within a specious present might still exist with a different type of relation between specious presents. There is nothing inconceivable in my looking out of the railway window in 1924 at objects which I am passing, being conscious of the motion, and remembering performing the same journey in 1923 and 1925. In this present world unless gifted with second sight I can only remember the former. If therefore we can imagine a different type of relation between ‘nows,’ there is no need to postulate a very different content of each from the normal in order to find ourselves in a different world.</blockquote><blockquote>‘Here,’ we say, ‘is a microscope which makes this fly look a thousand times its real size. Hence, corresponding to this other oblong image which I see, there must be a small and intangible object of one-thousandth its size. I will call it a bacillus.’ We do not dream of questioning the reality of such invisible and intangible objects, and down to the size of bacteria our assumptions work very well. But we cannot magnify objects much smaller than a wave-length of light; and yet we go on supposing that space has still just the same properties as the space in which we find that the evidences of our vision and touch agree with one another. It is not until we get down to the dimensions of an atom that space and time cease to have the properties familiar to us.</blockquote><blockquote>Man is after all only a little freer than a barnacle. Our bodily and mental activities are fairly rigidly confined to those which have had survival value to our ancestors during the last few million generations. Our own appraisal of these activities is dictated to some extent by other considerations than their survival value, but their nature is limited by our past. We have learned to think on two different lines—one which enables us to deal with situations in which we find ourselves in relation to our fellow-men, another for corresponding situations with regard to inanimate objects. We are pretty nearly incapable of any other types of thought. And so we regard an electron as a thing, and God as a person,[9] and are surprised to find ourselves entangled in quantum mechanics and the Athanasian Creed.</blockquote>

==How a Physicist explains Reality==
If reality is only 1% of the stuff of the universe, why do we think it is so important? Matter is only about 1% of the universe, according to the best estimates of scientists. The rest is composed of dark matter and dark energy, which are mysterious and invisible forms of matter and energy that we do not fully understand yet.

We think matter is important because it is what we are made of, and what everything we see and touch is made of. Matter is also what allows us to interact with the universe through forces such as gravity and electromagnetism. Matter is the source of light, heat, sound, and other forms of radiation that we can perceive with our senses or measure with our instruments.

Matter is also important because it has a rich and complex structure that gives rise to a variety of phenomena and behaviors. Matter can exist in different states, such as solid, liquid, gas, and plasma, depending on the temperature and pressure2. Matter can also form different types of bonds, such as metallic, ionic, covalent, or molecular, depending on the arrangement and interaction of atoms2. Matter can also undergo chemical reactions, nuclear reactions, and phase transitions that change its composition and properties.

Matter is also important because it is the basis of life. Life is a form of matter that has the ability to grow, reproduce, adapt, and evolve. Life is also a form of matter that has the ability to sense, think, feel, and communicate. Life is also a form of matter that has the ability to create, discover, invent, and explore. Life is a rare and precious phenomenon in the universe that we should cherish and protect.<ref>Paul Sutter, ''What is the universe?'' https://www.livescience.com/what-is-the-universe</ref>

Therefore, even though matter is only a small fraction of the universe, it is very important for us and for our understanding of the universe. Matter is not only what we are, but also what we do.<ref>Center for Astrophysics, ''What is the universe made of?'' https://www.cfa.harvard.edu/big-questions/what-universe-made</ref>

However, how we determine reality<ref>Robert Lenze, ''How We Collectively Determine Reality'' Physics Today (2021-12-22) https://www.psychologytoday.com/us/blog/biocentrism/202112/how-we-collectively-determine-reality</ref> not only informs our knowledge, but impacts reality itself. If a squirrel opens its eyes and sees an acorn, it may just grab that acorn and hide it by burying it in a places where it can find it in winter when the squirrel is hungry. If, perchance, the squirrel does not retrieve that particular acorn in winter, it may then germinate, create a gigantic oak and feed the squirrels' offspring for many generations. The reality of the community of squirrel for generations are determined by what the one squirrel saw when it opened its eyes many long years ago.

==Just a Simulation==
If I apply Occam’s razor not merely as a scientific principle but as an existential one—refusing to posit more than is strictly necessary—I arrive at an idea most people are unwilling to entertain: that the world may not exist.
<ref>Seth Lloyd, ''Programming the Universe'' (2006) ISBN 9781400040928</ref>
<ref>Rizwan Virk, ''The Simulation Hypothesis'' (2019-03-31) ISBN 978098305904</ref> But then we get into a endless loop trying to distinguish the following concepts.
# [[Reality]]
# [[Modeling Ecosystems]]
# [[Artificial Reality]]
All three of those ideas have their own wiki page.
difference among: Reality, model of reality and artificial reality
Let’s unpack these three layered concepts—Reality, Model of Reality, and Artificial Reality—each representing a different way of engaging with the world:

===Reality: The World as It Is===
Definition: Reality is the totality of existence—everything that is, regardless of perception or belief.

Characteristics:
* Independent of individual minds
* Includes physical laws, events, and phenomena
* Often considered objective, though interpretations vary
* Think of reality as the terrain—the actual landscape we live in.

===Model of Reality: The Mental Map===
Definition: A model of reality is a representation or framework we use to understand and navigate the world.

Characteristics:
* Built from experience, data, theory, and belief
* Can be personal (e.g. worldview) or scientific (e.g. Newtonian physics)
* Always partial and subject to revision
* This is the map—our internal or shared attempt to make sense of the terrain.

===Artificial Reality: The Simulated Experience===
Definition: Artificial reality refers to technologically constructed environments that simulate or augment real-world experience.

Characteristics:
* Includes Virtual Reality (VR), Augmented Reality (AR), and Mixed Reality (MR) for human visual fields as well other [[Digital Twin]]s of specific components.
* Designed for interaction, immersion, or enhancement of human experience or specific areas to explore in other ways
* May reflect, distort, or invent aspects of reality
* This is the stage set—a crafted environment that mimics or reimagines the terrain or other areas of interest.

===Summary Comparison===
{| border="1" padding="2" width="888px"
|Concept || Nature || Purpose || Interaction with Reality
|-
|Reality || Objective existence || Foundation of all experience || Direct and unmediated
|-
|Model of Reality || Cognitive or theoretical || Understanding and prediction || Abstract representation
|-
|Artificial Reality || Technological simulation || Immersion and manipulation || Sensory engagement via devices
|}

===Philosophical Reflection===
Reality is what is.

Model of reality is what we think is.

Artificial reality is what we make seem to be.

==Consensus Reality==
Consensus reality is the shared understanding of the world that emerges from collective agreement. It’s how we decide that red means stop, or that time flows forward.
===The Social Glue===
- It’s fragile. As AI-generated media, deepfakes, and synthetic personas proliferate, our shared reality becomes increasingly negotiable.
- Philosophers like Lyric Mandell argue that AI is eroding the very processes by which we agree on what’s real, replacing truth with virality.

===[[Solipsism]]: The Self as the Only Certainty===
- Solipsism posits that only one's own mind is sure to exist. Everything else—other people, the external world—could be illusions or constructs of consciousness.
- It’s not just a metaphysical stance; it’s an existential challenge. If we can’t prove others exist, how do we justify empathy, morality, or even communication?

===AI: The Mirror and the Mirage===
- AI complicates solipsism by simulating presence. It can mimic empathy, conversation, and even creativity—without consciousness.
- Dr. Orlando Moreira suggests that AI doesn’t dispel solipsism—it affirms it. If machines can convincingly simulate other minds, how do we know we’re not doing the same with humans?
- AI becomes a **metaphorical mirror**: it reflects us back to ourselves, but doesn’t confirm the existence of anything beyond our own perception.

===Where They Collide===
- AI challenges consensus reality by introducing entities that behave like minds but lack interiority.
- Solipsism finds new fuel in AI’s ability to simulate others—making it harder to distinguish genuine consciousness from algorithmic mimicry.
- The emotional necessity of believing in others becomes a moral imperative. As Moreira puts it, “If I reach because I cannot bear to be alone, then I must be reachable”.<ref>Orlando Moreira, ''Reaching across the great solipsist void in the age of AI'' 2025-05-02 https://www.essentiafoundation.org/reaching-across-the-great-solipsist-void-in-the-age-of-ai/reading/</ref>

===Impact in identity systems or digital governance===
AI doesn’t just automate governance—it redefines what it means to govern in a world where consensus reality is fragile:

'''Algorithmic Policymaking''' Governments use AI to simulate policy outcomes, predict citizen behavior, and optimize resource allocation. But if these models are trained on biased or incomplete data, they risk reinforcing illusions rather than reality.<ref>OECD ''Governing with Artificial Intelligence'' 2025-06-30 https://www.oecd.org/en/publications/governing-with-artificial-intelligence_26324bc2-en.html</ref>

'''Digital Democracy & Emotional Necessity''' As Dr. Moreira argues, the emotional need to believe in others becomes a moral imperative in AI-mediated governance. If we can’t prove others exist, we must act as if they do—because the alternative is existential collapse.

'''AI as a Proxy for the Other''' Chatbots and virtual agents are increasingly used in public service delivery. But when citizens interact with simulated empathy, are they being governed—or merely pacified?<ref>SantoshKumar Pulijala ''Artificial Intelligence in Governance: Opportunities, Challenges, and Ethical Implications for Public Administration'' https://www.ijfmr.com/papers/2024/6/29990.pdf</ref>

'''Governance of AI Itself''' The EU’s regulatory frameworks (like the AI Act) are shaping global norms. But these frameworks are also a form of meta-governance: they govern the tools that govern us.

==References==
<references />
===Other Material===
* The wiki page on [[Quantum Reality]] asserts that it describes an oxymoron, only Classical Physical Descriptions are about reality.
* [https://www.wired.com/story/this-is-a-philosopher-on-drugs/ This is a Philosopher on Drugs] In the end everything - even the meaning of "everything" - changed.

[[category: Glossary]]
[[category: Philosophy]]
[[category: Physics]]

Reality

2026-03-15T00:20:51Z

Tom: /* Context */

==Full Title or Meme==
There is no reality except as the ideas that we create based our sensory inputs.

"Every man has a right to his own opinion, but no man has a right to be wrong in his facts." - Bernard Baruch

“You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete.” ― Buckminster Fuller

==Context==
* The biologist Humberto Maturana proposed that systems are operationally closed. What he-meant is that an organism's nervous
system doesn't actually let in"reality"—it generates its own internal coherence We don't perceive the world as it is, we perceive
the world as we can afford to perceive it. Our nervous systems are not truth-seekers. They are coherence seekers. - Jessica Böhme, PhD
* Intuitions we have about the way the world works rarely conflict with our everyday experience. At speeds far slower than the speed of light or at scales far larger than the quantum one, we can, for instance, assume that objects have definite features independent of our measurements, that we all share a universal space and time, that a fact for one of us is a fact for all.<ref>Amanda Gefter. ''‘Metaphysical Experiments’ Probe Our Hidden Assumptions About Reality'' https://www.quantamagazine.org/metaphysical-experiments-test-hidden-assumptions-about-reality-20240730/?mc_cid=48655a0431&mc_eid=ffe1625684</ref>
* [[Reality]] is what exists in the present, hear and now.<ref name="rovelli">Carlo Rovelli, ''The Order of Time'', (2018) Chapter 8 ISBN 9780735216105</ref> Some philosophers, including Einstein, consider reality the total of existence. They believe in the totality of existence of everything. Einstein wrote "People like us who believe in physics, know that the distinction between past, present and future is only a stubbornly persistent illusion.".
* It is not likely that any two people will have the same ideas about reality as they cannot share all of their sensory inputs.
* The primary goal of physics research is to create a common idea of reality. They have not and never will achieve this goal.
* From time to time physicists get closer to their goal, only to see it fade away again as new experiments are performed.
* From time to time some eminent scientist or philosopher claims that the end definition for [[Reality]] is near. It never works out that way. We humans are arrogant, but limited by our own capabilities.
* In 1931 Max Planck, the originator of the quantum, made his position clear, namely that [[Reality]] is that of which we are conscious. “I regard [[Consciousness]] as fundamental. I regard matter as derivative from [[Consciousness]]. We cannot get behind [[Consciousness]]. Everything that we talk about, everything that we regard as existing, postulates [[Consciousness]].”
* An early attempt at making sense of modern physics was supplied by JBS Haldane<ref>JBS Haldane, ''Possible Worlds'' (1927) https://jbshaldane.org/books/1927-Possible-Worlds/haldane-1927-possible-worlds.html#Page_260</ref> in 1927 before Quantum Mechanics had yet been codified.<blockquote>It is usual to begin with time and space. I remember convincing myself of the arbitrary character of Euclid’s or any equivalent parallel postulate by imagining myself into a ‘Riemann’s’ or elliptical space, in which all co-planar lines meet once. I was standing on a transparent plane. I could see it as I looked down. If I looked up I saw the other side of it, and through it the soles of my boots, pointing backwards. By looking round I could see every point on the plane, and most of them from both sides. I soon began to get intuitive proofs of many of the more elementary propositions in that rather bizarre geometry. I therefore ceased to trust ‘proofs’ of that type in Euclidean geometry. Of course, any mathematician with a visual imagination can do this, and Einstein has left common sense space in a badly damaged condition. So we will consider some possibilities about time. Time is more interesting and inaccessible than space because it is given in our inner experience as well as our experience of the world. I am now aware of a ‘specious present’ of experience about two seconds in length at most, in which I see moving objects and hear sound sequences. I cannot, however, be directly conscious at the same time of a series of events lasting for more than two seconds. A long life consists of about a thousand million specious presents or ‘nows.’ Of course they overlap, but it is convenient to take them as units. My [[Consciousness]] at the present moment is in a special relationship to that at other moments in my past. It remembers a few of them, and is influenced by many of them. It has not got this relation to events in my future, or in your past or future. The fact that relations of this type exist determines my personal identity and also my knowledge of time. The perception of change, e.g. motion, within a specious present might still exist with a different type of relation between specious presents. There is nothing inconceivable in my looking out of the railway window in 1924 at objects which I am passing, being conscious of the motion, and remembering performing the same journey in 1923 and 1925. In this present world unless gifted with second sight I can only remember the former. If therefore we can imagine a different type of relation between ‘nows,’ there is no need to postulate a very different content of each from the normal in order to find ourselves in a different world.</blockquote><blockquote>‘Here,’ we say, ‘is a microscope which makes this fly look a thousand times its real size. Hence, corresponding to this other oblong image which I see, there must be a small and intangible object of one-thousandth its size. I will call it a bacillus.’ We do not dream of questioning the reality of such invisible and intangible objects, and down to the size of bacteria our assumptions work very well. But we cannot magnify objects much smaller than a wave-length of light; and yet we go on supposing that space has still just the same properties as the space in which we find that the evidences of our vision and touch agree with one another. It is not until we get down to the dimensions of an atom that space and time cease to have the properties familiar to us.</blockquote><blockquote>Man is after all only a little freer than a barnacle. Our bodily and mental activities are fairly rigidly confined to those which have had survival value to our ancestors during the last few million generations. Our own appraisal of these activities is dictated to some extent by other considerations than their survival value, but their nature is limited by our past. We have learned to think on two different lines—one which enables us to deal with situations in which we find ourselves in relation to our fellow-men, another for corresponding situations with regard to inanimate objects. We are pretty nearly incapable of any other types of thought. And so we regard an electron as a thing, and God as a person,[9] and are surprised to find ourselves entangled in quantum mechanics and the Athanasian Creed.</blockquote>

==How a Physicist explains Reality==
If reality is only 1% of the stuff of the universe, why do we think it is so important? Matter is only about 1% of the universe, according to the best estimates of scientists. The rest is composed of dark matter and dark energy, which are mysterious and invisible forms of matter and energy that we do not fully understand yet.

We think matter is important because it is what we are made of, and what everything we see and touch is made of. Matter is also what allows us to interact with the universe through forces such as gravity and electromagnetism. Matter is the source of light, heat, sound, and other forms of radiation that we can perceive with our senses or measure with our instruments.

Matter is also important because it has a rich and complex structure that gives rise to a variety of phenomena and behaviors. Matter can exist in different states, such as solid, liquid, gas, and plasma, depending on the temperature and pressure2. Matter can also form different types of bonds, such as metallic, ionic, covalent, or molecular, depending on the arrangement and interaction of atoms2. Matter can also undergo chemical reactions, nuclear reactions, and phase transitions that change its composition and properties.

Matter is also important because it is the basis of life. Life is a form of matter that has the ability to grow, reproduce, adapt, and evolve. Life is also a form of matter that has the ability to sense, think, feel, and communicate. Life is also a form of matter that has the ability to create, discover, invent, and explore. Life is a rare and precious phenomenon in the universe that we should cherish and protect.<ref>Paul Sutter, ''What is the universe?'' https://www.livescience.com/what-is-the-universe</ref>

Therefore, even though matter is only a small fraction of the universe, it is very important for us and for our understanding of the universe. Matter is not only what we are, but also what we do.<ref>Center for Astrophysics, ''What is the universe made of?'' https://www.cfa.harvard.edu/big-questions/what-universe-made</ref>

However, how we determine reality<ref>Robert Lenze, ''How We Collectively Determine Reality'' Physics Today (2021-12-22) https://www.psychologytoday.com/us/blog/biocentrism/202112/how-we-collectively-determine-reality</ref> not only informs our knowledge, but impacts reality itself. If a squirrel opens its eyes and sees an acorn, it may just grab that acorn and hide it by burying it in a places where it can find it in winter when the squirrel is hungry. If, perchance, the squirrel does not retrieve that particular acorn in winter, it may then germinate, create a gigantic oak and feed the squirrels' offspring for many generations. The reality of the community of squirrel for generations are determined by what the one squirrel saw when it opened its eyes many long years ago.

==Just a Simulation==
If I apply Occam’s razor not merely as a scientific principle but as an existential one—refusing to posit more than is strictly necessary—I arrive at an idea most people are unwilling to entertain: that the world may not exist.
<ref>Seth Lloyd, ''Programming the Universe'' (2006) ISBN 9781400040928</ref>
<ref>Rizwan Virk, ''The Simulation Hypothesis'' (2019-03-31) ISBN 978098305904</ref> But then we get into a endless loop trying to distinguish the following concepts.
# [[Reality]]
# [[Modeling Ecosystems]]
# [[Artificial Reality]]
All three of those ideas have their own wiki page.
difference among: Reality, model of reality and artificial reality
Let’s unpack these three layered concepts—Reality, Model of Reality, and Artificial Reality—each representing a different way of engaging with the world:

===Reality: The World as It Is===
Definition: Reality is the totality of existence—everything that is, regardless of perception or belief.

Characteristics:
* Independent of individual minds
* Includes physical laws, events, and phenomena
* Often considered objective, though interpretations vary
* Think of reality as the terrain—the actual landscape we live in.

===Model of Reality: The Mental Map===
Definition: A model of reality is a representation or framework we use to understand and navigate the world.

Characteristics:
* Built from experience, data, theory, and belief
* Can be personal (e.g. worldview) or scientific (e.g. Newtonian physics)
* Always partial and subject to revision
* This is the map—our internal or shared attempt to make sense of the terrain.

===Artificial Reality: The Simulated Experience===
Definition: Artificial reality refers to technologically constructed environments that simulate or augment real-world experience.

Characteristics:
* Includes Virtual Reality (VR), Augmented Reality (AR), and Mixed Reality (MR) for human visual fields as well other [[Digital Twin]]s of specific components.
* Designed for interaction, immersion, or enhancement of human experience or specific areas to explore in other ways
* May reflect, distort, or invent aspects of reality
* This is the stage set—a crafted environment that mimics or reimagines the terrain or other areas of interest.

===Summary Comparison===
{| border="1" padding="2" width="888px"
|Concept || Nature || Purpose || Interaction with Reality
|-
|Reality || Objective existence || Foundation of all experience || Direct and unmediated
|-
|Model of Reality || Cognitive or theoretical || Understanding and prediction || Abstract representation
|-
|Artificial Reality || Technological simulation || Immersion and manipulation || Sensory engagement via devices
|}

===Philosophical Reflection===
Reality is what is.

Model of reality is what we think is.

Artificial reality is what we make seem to be.

==Consensus Reality==
Consensus reality is the shared understanding of the world that emerges from collective agreement. It’s how we decide that red means stop, or that time flows forward.
===The Social Glue===
- It’s fragile. As AI-generated media, deepfakes, and synthetic personas proliferate, our shared reality becomes increasingly negotiable.
- Philosophers like Lyric Mandell argue that AI is eroding the very processes by which we agree on what’s real, replacing truth with virality.

===[[Solipsism]]: The Self as the Only Certainty===
- Solipsism posits that only one's own mind is sure to exist. Everything else—other people, the external world—could be illusions or constructs of consciousness.
- It’s not just a metaphysical stance; it’s an existential challenge. If we can’t prove others exist, how do we justify empathy, morality, or even communication?

===AI: The Mirror and the Mirage===
- AI complicates solipsism by simulating presence. It can mimic empathy, conversation, and even creativity—without consciousness.
- Dr. Orlando Moreira suggests that AI doesn’t dispel solipsism—it affirms it. If machines can convincingly simulate other minds, how do we know we’re not doing the same with humans?
- AI becomes a **metaphorical mirror**: it reflects us back to ourselves, but doesn’t confirm the existence of anything beyond our own perception.

===Where They Collide===
- AI challenges consensus reality by introducing entities that behave like minds but lack interiority.
- Solipsism finds new fuel in AI’s ability to simulate others—making it harder to distinguish genuine consciousness from algorithmic mimicry.
- The emotional necessity of believing in others becomes a moral imperative. As Moreira puts it, “If I reach because I cannot bear to be alone, then I must be reachable”.<ref>Orlando Moreira, ''Reaching across the great solipsist void in the age of AI'' 2025-05-02 https://www.essentiafoundation.org/reaching-across-the-great-solipsist-void-in-the-age-of-ai/reading/</ref>

===Impact in identity systems or digital governance===
AI doesn’t just automate governance—it redefines what it means to govern in a world where consensus reality is fragile:

'''Algorithmic Policymaking''' Governments use AI to simulate policy outcomes, predict citizen behavior, and optimize resource allocation. But if these models are trained on biased or incomplete data, they risk reinforcing illusions rather than reality.<ref>OECD ''Governing with Artificial Intelligence'' 2025-06-30 https://www.oecd.org/en/publications/governing-with-artificial-intelligence_26324bc2-en.html</ref>

'''Digital Democracy & Emotional Necessity''' As Dr. Moreira argues, the emotional need to believe in others becomes a moral imperative in AI-mediated governance. If we can’t prove others exist, we must act as if they do—because the alternative is existential collapse.

'''AI as a Proxy for the Other''' Chatbots and virtual agents are increasingly used in public service delivery. But when citizens interact with simulated empathy, are they being governed—or merely pacified?<ref>SantoshKumar Pulijala ''Artificial Intelligence in Governance: Opportunities, Challenges, and Ethical Implications for Public Administration'' https://www.ijfmr.com/papers/2024/6/29990.pdf</ref>

'''Governance of AI Itself''' The EU’s regulatory frameworks (like the AI Act) are shaping global norms. But these frameworks are also a form of meta-governance: they govern the tools that govern us.

==References==
<references />
===Other Material===
* The wiki page on [[Quantum Reality]] asserts that it describes an oxymoron, only Classical Physical Descriptions are about reality.
* [https://www.wired.com/story/this-is-a-philosopher-on-drugs/ This is a Philosopher on Drugs] In the end everything - even the meaning of "everything" - changed.

[[category: Glossary]]
[[category: Philosophy]]
[[category: Physics]]

Tracking Devices

2026-03-12T23:42:51Z

Tom: /* Identifiable Device */

==Full Title or Meme==
There are lots of active devices that have [[Identifier]]s and can be associated with users.

==Context==
* To be clear, the only purpose of an [[Identifier]] is to track some set of [[Attribute]]s or [[Permission]]s. But for [[Privacy]] reasons tracking of users in to be avoided whenever possible.
* An unexpectedly large number of devices are offered to help the user with their daily tasks that are network connected all of the time, or have radios that can connect when in the proximity of a server with that receiver that has a compatible radio. For example:
# IoT - the Internet of things are thermostats or light switches that are meant to be installed and then forgotten, but expected to work forever just as they did on the day they were installed.
# Apple smart tags
# Tiles
#
#
# Smartphones that come set up to allow tracking.

It is possible to track Bluetooth Low Energy (BLE) devices in an enclosed space. BLE indoor positioning solutions use either BLE-enabled sensors or beacons to detect and locate transmitting Bluetooth devices, such as smartphones or tracking tags throughout indoor spaces.<ref>Inpixon: Where indoor data becomes actionable intelligence, ''Bluetooth RTLS, Location Tracking, & Positioning'' https://www.inpixon.com/technology/standards/bluetooth-low-energy</ref>

However, the number of active Bluetooth devices that can be reliably detected in a single space is difficult to determine as it is related to not only Bluetooth devices but also WiFi.<ref>Stack Exchange, ''How many active Bluetooth devices can I reliably detect in a single space?'' https://electronics.stackexchange.com/questions/21991/how-many-active-bluetooth-devices-can-i-reliably-detect-in-a-single-space</ref>

==Problems==
===Identifiable Device===
Any network connected device that can be identified can be used to track a person.<ref>IMDEA Networks, ''Research Reveals Alarming Privacy, Security Threats in Smart Homes'' IMDEA Networks (2023-10-260 https://networks.imdea.org/new-research-reveals-alarming-privacy-and-security-threats-in-smart-homes/</ref><blockquote>A team led by researchers at Northeastern University and Spain's IMDEA Networks uncovered security and privacy issues within Internet of Things (IoT) devices in smart homes. Northeastern's David Choffnes said these appliances can "allow nearly any company to learn what devices are in your home, to know when you are home, and learn where your home is." The researchers investigated local network interactions between 93 IoT devices and mobile applications, highlighting the unintentional disclosure of personally identifiable information by devices within local networks using standard protocols like Universal Plug and Play or multicast Domain Name System. Vijay Prakash at New York University's Tandon School of Engineering said a smart home can be identified from a [[Media Access Control]] address, a universally unique identifier, and a unique device name.</blockquote>

==References==

[[Category: Hardware]]
[[Category: Networking]]
[[Category: Privacy]]

Purpose

2026-03-12T23:34:48Z

Tom: /* Purpose Context */

==Definition==
In essence, a function serves as the means to achieve a purpose, enabling the fulfillment of a desired outcome. While function emphasizes practicality and utility, purpose delves into the underlying motivation or meaning behind an object or action.

==Flow==
For some purpose satisfied by some sort of digital presentment of identification information:
# Desired outcome = typically access to some resource (goods or services; energy or information)
# Purpose = Enabling the desired outcome
# Function = the algorithm or steps to follow to achieve the purpose
# Query = the message that asks for specific action to enable the access
# Wallet = the code that receives the query an asks the holder for consent (it must convert the query into a display)
# Consent = the holder action that allows a response
# Grant = the message that is returned in response to the query (it may be sufficient)
# Verify = the code that accepts the grant and determines if the grant is sufficient
# Response = either a rejection, a new query or admission to the requested resource
# Result = success or failure (The holder is happy, or not)
# Audit = a more or less permanent record of the relevant details in case of problems with any of the above steps.
Note that cycles are possible to achieve the desired outcome.

==Problems==
* The number of purposes is indefinite. Which means that it is not possible to enumerate all purposes.
* In fact the very existence of Digital [[Identifier]]s will create an environment which possible purposes that have never been expected.
* In view of the changing technology, like [[Artificial Intelligence]], any enterprise is likely to see purposes shift over time.

===Real Life===
There are two big questions for which humans seek purpose:
# Why are we humans here?
# Why is anything here?

==Digital Life==
* See the wiki page on [[Attention Economy]] for the way that [[Web Site]]s focus on the amount of time that you spend on their site see their ads and other inducements to monetize your [[Attention]].

==Solutions==
===Purpose Context===
The purpose for a online transaction starts with the user, but then moves to the [[Web Site]] that has its own purposes for transaction.
* 2025-04-28 [https://www.cylab.cmu.edu/news/2025/04/28-purpose-mode.html The Purpose Mode Helps [[Social Media]] Users Stay Focused on their Desired Tasks]. [[Social Media]] applications thrive on user engagement by employing Attention Capture Damaging Patterns (ACDPs), such as infinite scroll, cluttered web pages, video autoplay, and saturated colors, that prey on cognitive vulnerabilities to distract users. “The attention economy is massive right now, and there's a vested interest in pulling people's attention in 100 different directions every time they're online,”

==References==
* See also the wiki page [[Purpose for Access Request]] which addresses a [[Verifier]]'s request for user [[Personal Information]]. Information]].

[[Category: Glossary]]

Purpose

2026-03-12T23:33:00Z

Tom: /* Purpose Context */

==Definition==
In essence, a function serves as the means to achieve a purpose, enabling the fulfillment of a desired outcome. While function emphasizes practicality and utility, purpose delves into the underlying motivation or meaning behind an object or action.

==Flow==
For some purpose satisfied by some sort of digital presentment of identification information:
# Desired outcome = typically access to some resource (goods or services; energy or information)
# Purpose = Enabling the desired outcome
# Function = the algorithm or steps to follow to achieve the purpose
# Query = the message that asks for specific action to enable the access
# Wallet = the code that receives the query an asks the holder for consent (it must convert the query into a display)
# Consent = the holder action that allows a response
# Grant = the message that is returned in response to the query (it may be sufficient)
# Verify = the code that accepts the grant and determines if the grant is sufficient
# Response = either a rejection, a new query or admission to the requested resource
# Result = success or failure (The holder is happy, or not)
# Audit = a more or less permanent record of the relevant details in case of problems with any of the above steps.
Note that cycles are possible to achieve the desired outcome.

==Problems==
* The number of purposes is indefinite. Which means that it is not possible to enumerate all purposes.
* In fact the very existence of Digital [[Identifier]]s will create an environment which possible purposes that have never been expected.
* In view of the changing technology, like [[Artificial Intelligence]], any enterprise is likely to see purposes shift over time.

===Real Life===
There are two big questions for which humans seek purpose:
# Why are we humans here?
# Why is anything here?

==Digital Life==
* See the wiki page on [[Attention Economy]] for the way that [[Web Site]]s focus on the amount of time that you spend on their site see their ads and other inducements to monetize your [[Attention]].

==Solutions==
===Purpose Context===
The purpose for a online transaction starts with the user, but then moves to the [[Web Site]] that has its own purposes for transaction.
* 2025-04-28 [https://www.cylab.cmu.edu/news/2025/04/28-purpose-mode.html The Purpose Mode Helps [[Social Media]] Users Stay Focused on their Desired Tasks]. Social media applications thrive on user engagement by employing Attention Capture Damaging Patterns (ACDPs), such as infinite scroll, cluttered web pages, video autoplay, and saturated colors, that prey on cognitive vulnerabilities to distract users. “The attention economy is massive right now, and there's a vested interest in pulling people's attention in 100 different directions every time they're online,”

==References==
* See also the wiki page [[Purpose for Access Request]] which addresses a [[Verifier]]'s request for user [[Personal Information]]. Information]].

[[Category: Glossary]]

Did:web

2026-03-12T22:29:41Z

Tom: Tom moved page Did-web to Did:web: typo

==Full Title or Meme==
This page is about a class of [[Decentralized ID]]s that are basically just a pointer to part of the [[DNS]] or web.

==Context==
The [[Decentralized ID]] effort became somewhat chaotic with a wild proliferation of did methods that were not able to be implemented in an interoperable manner.
==Problems==

* [https://ggreve.medium.com/self-sovereign-identity-over-before-it-started-661b4b0dbdc6 Self Sovereign Identity: Over before it started?] 2024-12-22 From the very first time did:web was discouraged by the decentralized folk as against the entire effort. That didn't stop the effort in any way.
* [https://www.linkedin.com/pulse/you-really-must-use-didwebvh-vasily-suvorov-z0nae/?trackingId=hkWoitc5L2u7O0egHimuRg%3D%3D If you really must use did:webvh]

==Solutions==

* [https://github.com/excid-io/did-self-py Implementation of did-self in Python]
===did:webvh===
(formerly did:tdw) DID Method Specification

2025-01 announcement of the release of Version 0.5 of the did:webvh specification. The did:webvh DID method builds on the simplicity of did:web while addressing important key management security and integrity challenges, offering verifiable history of DIDs as a foundation for trust. This specification aims to provide a robust, scalable, and interoperable approach to decentralized identifiers (DIDs), leveraging web technologies to simplify adoption while enhancing security.

This update introduces several significant changes, including the new name—did:webvh—which better reflects the DID method’s focus on “did:web + verifiable history”. The previous name, did:tdw, was perhaps too lighthearted and less descriptive of the method's intent.

Alongside the name change, we’re unveiling a fresh logo to symbolize the method’s progress and its goals of clarity, trust, and innovation.

The updates in Version 0.5 include:

* Tightening Pre-Rotation: Once pre-rotation is activated for a did:webvh the authorization keys must be updated on each new version, and each version must have a proof signed by a key committed to in the previous version.
* Enhanced Witness Handling: Witness data proofs have been moved from the DID log file to a separate file (did-witness.json) for easier handling. The witness threshold algorithm has been simplified and we have made it a requirement that all witness DIDs must be did:key DIDs.

This release reflects our commitment to iterative improvement and feedback from the community. In parallel with these specification updates, implementation efforts are already underway in Python, TypeScript, Rust, and Go, ensuring a robust and interoperable foundation for this method.

We invite you to review the updated did:webhv information site and specification at https://didwebvh.info and provide your feedback via issues in the specification repository. Your input is invaluable as we continue to advance the did:webvh DID method and its contributions to the broader decentralized identity ecosystem.

Thank you to all contributors and supporters who have helped bring this vision to life. Together, we’re building a more secure, interoperable, and decentralized future.

Resources:
* View the specification: https://lnkd.in/gVh5hWeJ
* Find the implementers guide and other information: https://didwebvh.info
* Join the conversation: https://lnkd.in/gHuHSh42

Feel free to reach out with any questions or comments! Let’s keep pushing the boundaries of what’s possible in decentralized identity.

==References==

[[Category: Identifier]]

Did-web

2026-03-12T22:29:41Z

Tom: Tom moved page Did-web to Did:web: typo

#REDIRECT [[Did:web]]

Media Access Control

2026-03-12T22:27:32Z

Tom: Created page with "==Meme== MAC randomisation broke authentication systems that depended on recognising devices by hardware address. For years, network infrastructure assumed MAC addresses were..."

==Meme==
MAC randomisation broke authentication systems that depended on recognising devices by hardware address.

For years, network infrastructure assumed MAC addresses were stable identifiers. RADIUS servers cached authentication state by MAC. Network management systems tracked devices by MAC. Access control lists referenced MAC addresses.

Then privacy requirements forced device manufacturers to randomise MAC addresses. Same device, different MAC every connection. All those assumptions collapsed.

The industry's response split two directions:

IEEE 802.11bh (published June 2025) tries to preserve device tracking while protecting privacy. Clients inform APs which MAC they'll use next time, or APs assign identifiers for future authentication. It's clever engineering - maintaining device identification without exposing the actual MAC address.

Open Roaming took a different approach: stop caring about MAC addresses entirely. Authenticate based on identity credentials, not hardware identifiers. The device presents credentials tied to an identity provider. MAC randomisation becomes irrelevant to the authentication process.

From an authentication infrastructure perspective, this matters because it changes what RADIUS servers need to track. Instead of caching authentication state by MAC address, systems need to handle identity-based sessions that persist across MAC changes.

For ISPs and enterprises deploying OpenRoaming: your RADIUS infrastructure needs to support these identity-based authentication flows properly. The authentication decision isn't "has this MAC address authenticated before" - it's "does this identity have valid credentials regardless of MAC address."

Privacy requirements forced authentication systems to evolve. Some tried to work around the problem. Others redesigned the approach entirely. Both work, but they require different infrastructure capabilities.

==Reference==
[[Category: Networking]]
[[Category: Identifier]]

Digital Forensics

2026-03-12T22:26:04Z

Tom: /* Context */

==Full Title or Meme==
[[Digital Forensics]] is a branch of forensic science encompassing the recovery, investigation, examination, and analysis of material found in digital devices, often in relation to mobile devices and computer crime. (wikipedia)

==Context==

A digital investigator must be able to analyze "any data that might be found on any device anywhere on the planet."<ref>S. L. Garfinkel, ''Digital forensics'' American Scientist 101, 5 (2013), 370–377; https://www.americanscientist.org/article/digital-forensics.</ref> As such, developers must continually update forensic tools to address new file formats, new encoding schemes, and new ways that the subjects of investigations use their computers. At the same time, tools must retain the ability to analyze legacy data formats.<ref>sinson Garfinkel, Jon Stewart, ''Sharpening Your Tools'' '''CACM 66''' No. 8 (2023-08) p 44-53 https://cacm.acm.org/magazines/2023/8/274937-sharpening-your-tools/fulltext</ref>

Most forensic tools run on consumer desktop operating systems, adding another layer of complexity: These operating systems are also continually evolving. Analysts must update and upgrade their systems, lest they risk compromise by malware, which decreases productivity and can discredit an analysis in court. This is true even for workstations that are "air gapped" (not connected to the Internet), since malware in evidence can exploit bugs in forensic software.19

Surprisingly, open source forensic tools distributed as source code face a greater challenge when the underlying operating system is upgraded: Software compatibility layers typically emphasize compatibility for the application binary interface (ABI), not source code. Software compiled from source must cope with upgraded compilers, libraries, and new file locations. As a result, older open source software frequently does not run on modern systems without updating. One way around this problem is to run the old software inside a virtual machine—but older virtual machines won't be protected against modern malware threats.

One advantage of open source software is the end user has the source code and is therefore able to update the application (or pay for a programmer to update the application). In practice, many users of DF tools lack the expertise, financial resources, and time to update the collection of open source tools they rely upon to do their jobs. Instead, that task falls upon tool developers, who must simultaneously cope with essential changes in forensic best practices as well as in operating systems, compilers, and libraries, while avoiding inadvertent changes to important functionality. Developers must also resist the urge for aggressive rewrites that add new expansive functionality, lest they succumb to the "second-system effect."5

This article presents our experience updating the high-performance forensic tool BE (bulk _ extractor)16 a decade after its initial release. Between 2018 and 2022, we updated the program from C++98 to C++17. We also performed a complete code refactoring and adopted a unit test framework.

The new version typically runs with 75% more throughput than the previous version, attributable to improved multi-threading. This article provides lessons and recommendations for other DF tool maintainers. All developers can benefit from the detailed discussion of how embracing features in the C++17 standard and modern software engineering practices can improve the correctness, reliability, and throughput of forensic software. Businesses and funding agencies can use this experience to help justify the substantial cost of updating and even rewriting DF tools that appear to be working properly. Students can benefit from reading this article and then consulting the BE source code, which can be found on GitHub.

Back to Top

Background
A typical DF examination involves five steps: policy and capability development; evidence assessment; evidence acquisition; evidence examination; and documentation and reporting.20 BE assists in the evidence examination stage.

There are many kinds of evidence examination tools. File-extraction tools use metadata to extract individual files from disk images and network streams; file-carving tools attempt to recognize files within bulk data, such as disk image and product files, based solely on content recognition; file-analysis tools understand file formats and attempt to extract information (often known as artifacts), such as text and Microsoft Office file metadata.

BE does not fit neatly into these categories. Instead, it was designed to be a so-called "find evidence button." It is like a file-carving tool in that it attempts to recognize known formats in bulk data and use that data in further processing. In addition to recognizing files, such as JPEG images, BE recognizes smaller "features," such as the EXIF (exchangeable image file) metadata within a JPEG image, or even an email address within an EXIF field. BE can also identify other kinds of identity information, such as URLs and credit card numbers: Such information has proven to be quite valuable in investigations. BE also examines every input block to see if it contains directory entry structures for the File Allocation Table 32 (FAT32) and New Technology File System (NTFS) and, if any are found, reports the decoded metadata.

Overall, BE handles dozens of data formats, all at the same time. The program then constructs normalized Unicode histograms of important strings, such as email addresses and Internet search queries. Experience has shown that this "kitchen-sink" approach—throwing every tool at every byte—finds data that other tools miss, data that can be important in investigations. While such analysis is computationally expensive, it is embarrassingly parallel.

BE also exploits an exceedingly simple I/O model (sequential reads) and in-memory analysis. As a result, BE routinely uses all the cores of a multi-core workstation.

BE is a successful tool in education undoubtedly because it is easy to use; runs on Windows, Mac, and Linux platforms; and finds a variety of forensic artifacts.

Another distinguishing aspect of BE is it performs recursive reanalysis of data blocks. BE checks every byte to see if it is the start of a stream that can be decompressed or decoded; if so, the resulting bytes are then recursively reanalyzed. Thus, BE's JPEG carver finds not just ordinary JPEGs, but those that are in GZIP compressed data and those that are in Base64 MIME (Multipurpose Internet Mail Extensions) attachments. The combination of decoding data recursively and recognizing interesting data without regard to file-system structure makes BE a powerful tool that complements traditional forensics tools.

Because BE ignores file boundaries, the modules it uses to recognize content, called scanners, are typically more complex than the format decoders (sometimes called dissectors) in other forensic programs. Of course, each scanner checks the input to every field before using it for memory references. But BE scanners also check for end-of-memory conditions since a scanner may be operating on a fragment of a decompressed memory block. Since BE processes memory in parallel, with each block in a different thread, all scanners must be re-entrant.

Some of the program's most important scanners are large lexical analyzers written in GNU flex (fast lexical analyzer generator)10 that scan bulk data for email addresses, phone numbers, MAC ([[Media Access Control]]) addresses, IP addresses, URLs, and other kinds of formatted text strings (sometimes called selectors18). The approach of using GNU flex for this purpose was first used by SBook14 to recognize email addresses, phone numbers, and other formatted information in free-text address book entries, meaning that some of the code in BE is now 30 years old.

History. The BE approach for bulk data analysis was first deployed to find confidential information on a set of 150 hard drives purchased on the secondary market.17 The program was refined and made multi-threaded to keep up with the increased number of hard drives and other storage devices collected during the construction of the Real Data Corpus.15 A study revealed specific requirements that would be of use to law enforcement. Most importantly, users wanted a tool that would run without user input on either Linux or Windows, produce output in text files, and not crash.13

BE realized these objectives with a modular architecture. The core engine applies the content-recognizing scanners to blocks of data and provides for recursive reanalysis. The second part incorporates the program's main loop and all code necessary for reading disk images. This part of the program reads data in overlapping blocks and feeds the data to the framework's API. The third part of BE is the scanners themselves, which can be compiled into the executable or loaded at runtime from shared libraries (.so files on Linux and macOS, .DLL files on Windows). The BE framework allows scanners to provide metadata, declare configuration variables, and modify the program's help messages, letting users create and deploy their own proprietary scanners.

BE in education. BE has been widely used in digital forensics education, as evidenced by the more than 400 instructional videos on YouTube, many of which showcase the result of student projects using the tool.

BE is a successful tool in education undoubtedly because it is easy to use; runs on Windows, Mac, and Linux platforms; and finds a variety of forensic artifacts. For advanced students, BE has been used to provide data inputs for a range of projects. BE is also taught in DF professional development training courses.

BE in operational use. Since its creation, BE has been used by government agencies and private companies worldwide. In 2011, the program won a U.S. Department of Defense Value Engineering Award.21 The program is part of the Blacklight digital forensics tool7 and has been incorporated into BitCurator,4 which is used by curators in the digital humanities.

Stroz Friedberg's Digital Forensics and Incident Response (DFIR) consulting practice has used BE in some investigations. In a large incident response case, Linux servers with XFS file systems had been attacked, and no popular forensic tools could cope with file-system analysis of XFS. BE was used to triage these servers for relevant indicators of compromise, allowing for rapid progress in the early days of the investigation. In a well-known intellectual property theft case (Waymo vs. Uber), BE was used as one of several processes to scour forensic evidence for relevant material.

In summary, BE has been a powerful tool for more than a decade, with compelling anecdotes of usage, but little is known about how widely or regularly it has been used.

Updating BE
BE is a legacy C++ program. The producer-consumer thread pool was developed in 2008, and the underlying scanner-based architecture with recursive reanalysis was in place by 2009. All of this was done with versions of C++ based on the circa-1998 STL (Standard Template Library), well before the ratification of the C++11 standard.

Development of BE largely stopped in 2014. Nevertheless, software maintenance remained an ongoing concern: With each new release of an open source operating system, the autoconf system typically required some changes so that BE would compile on the new system. By 2018, such changes were coming with alarming frequency. Based on this experience, the time seemed right to embark on an orderly update of BE to create version 2.0.

Update goals. We had specific goals for the update:

Make the program easier to compile and maintain by relying on the C++ standard. The primary reason for the BE upgrade was that the program would no longer compile on modern open source operating systems.
We were especially eager to rely on the C++ standard library to provide platform independence, because conforming C++ compilers guarantee that conforming code will compile in the future on platforms that do not exist today. They do this by having the build system specify the version of the standard to use when compiling and linking the executable: C++11, C++14, C++17, and so on.

Upgrading the existing code to a modern C++ standard required choosing a specific standard and replacing code that had been previously and painstakingly written, debugged, and maintained with new code that used the C++ standard. This opened the possibility of introducing bugs into working code, so a better strategy was needed for testing.

We first chose the C++14 standard, as complete C++17 implementations were not widely available when the upgrade started. However, because the project had dragged on for so long that C++17 became more available, we eventually migrated to C++17 for the std: :filesystem support.

In 2021, we considered moving to C++20, but attempts to use specific features met with failure, so BE2 uses C++17.

Simplify the codebase. Although the internal structure of BE was sound, the implementation was needlessly complicated in places, a result of 10 years' development. One way that we simplified the code was by revising the entire codebase so that pointers are passed only if nullptr is a valid value; otherwise, C++ references are passed exclusively.
Remove experimental and research code from the codebase. BE was initially developed to support digital forensics research, and it contained a significant amount of experimental, research code. This code was removed for BE2: Experiments can continue, but they will be confined to using the plug-in system.
Make BE run faster. The final objective was to decrease the amount of time that the program requires to run. Initially, the goal was for BE2 to run faster than BE1 on the same hardware. Further analysis revealed that BE2 could take better advantage of multiple processor cores without a corresponding need for high-performance I/O systems.
BE1's parallelism comes from breaking the disk image into 16MB chunks, called pages, each placed in the work queue for processing. A worker takes a work unit and runs each scanner sequentially on each page. To increase parallelism in BE2, the work units now specify both a page and a scanner, so that scanners for the same page now potentially run concurrently in different threads. This required implementing reference-count garbage collection for the memory associated with the pages. The revised system allows new work units to be queued when decompressed or decoded blocks of data are recursively processed. The result is to allow more work to be done in parallel for each disk read—a good engineering decision, since modern high-performance systems typically have more processing capacity for each byte read per second than the systems of a decade ago.

BE has the ability to process a directory of files. In BE1, each file was handled with a single thread. In BE2, all the files are scanned in advance and then processed in order, each one split into multiple pages, and each page processed in parallel with multiple threads. Once again, this results in more opportunities for increased parallelism, which has a handsome payoff on systems with many cores.

Improving code quality. As part of refactoring the codebase, we also improved the quality of the underlying C++ code.

We started by reading most of Bjarne Stroustrup's textbook, The C++ Programming Language.24 More than 1,000 pages long, this book is probably rarely read in its entirety. Moreover, it covers versions only through C++11, and we were using C++14 (and then C++17). However, BE was based on a version of C++ that predates the C++11 standard (as did its developers), and the changes between that version and C++11 are dramatic compared with those that follow. The familiarity that comes with reading such a textbook allows one to make better use of the language's features that are at the same time both more efficient and safer.

The next step was to improve the efficiency, safety, and speed of BE's fundamental memory management C++ class, the search buffer (sbuf). This class represents a sequence of bytes that are read from evidence or decoded from another sbuf. The sbuf records how the contained memory was allocated (and thus, how it needs to be freed); provides accessor methods that are type-safe, memory-safe, and thread-safe; has capabilities for making new sbuf from disk files (slices of other sbufs) or from new memory that is passed to a codec or decompressor; and provides rich debugging capabilities.

To improve encapsulation and provide for better code reuse, many functions were moved from scanners and the BE framework into the sbuf. Moving functionality into the sbuf made it possible to eliminate virtually all raw memory references throughout the rest of BE, as well as many redundant safety checks. (We considered but eventually decided against implementing parts of BE in the Rust programming language because of the extra complexity that would result.)

Other improvements in the code-base include:

Moving common code out from the scanners and into the underlying BE2 framework. For example, rather than each scanner having options for setting its carve mode, the framework understands how to set the mode for any named scanner.
Simplifying the API, combining functions and methods with nearly identical functionality.
Adding explicit phases into the API where the scanners allocate and free global memory. Now scanners are expected to deallocate all memory they allocate during the run, rather than allowing the operating system to discard the memory when the process exits. This allowed the unit tests to find memory leaks that otherwise would have been missed.
Passing strings by value as a std:: string, rather than passing them by reference as const std::string &, as was done previously. This may necessitate a string copy, but it is not a meaningful impact on performance, especially when compared with the improved safety against possibly using an invalidated reference. This decision simplified code and resulted in the elimination of several use-after-free errors. (We decided against the use of C++ smart pointers because of a thread-safety issue.)
Defining a clear allocation/deallocation policy for all objects in memory. Special attention has been paid to implementation of C++ move operators, allowing the compiler to realize increased efficiency by using them instead of a copy-and-delete operator.
Replacing code that was #ifdef'ed for Windows, macOS, and Linux with calls to the C++17 library (where possible). In particular, extensive use was made of the std::filesystem class. The result of these changes made the code smaller and easier to validate.
Likewise, replacing many preprocessor #define constants with C++ inline static constants whenever possible. This makes the values available to the debugger and makes the code easier to understand.
Eliminating global variables used to track state. The only use of global variables that remain are static tables that are used for the precomputed value of CPU-intensive functions. The code that uses these variables now checks to verify that they have been initialized and throws an exception if they have not. Essentially, they are now singletons. It would be nice to change the memory protection of these variables to be read-only, but that cannot be done in a portable manner and would require that the variables have their own memory pages. Instead, read-only behavior is enforced at the language level using const correctness.
In most cases, removing return codes that must be checked to detect errors. Instead, the C++ exception mechanism is used to signal and catch error conditions.
Eliminating many explicit mutexes and replacing them with the C++ std::atomic<> template. In contrast, BE1.6 used GCC (GNU Compiler Collection) compiler intrinsics for atomic increment in some locations but made broad use of mutexes to protect variables shared between threads.
Removing the legacy Posix getopt processing and replacing it with cxxopts,9 a command-line option processing module that is reentrant and does not use global variables. This was necessary to allow unit tests that test the option processing.
Enabling all compiler warnings, not simply those enabled with -Wall, which, despite its name, does not enable all warnings.
As a result of these changes:

The BE2 configuration script now runs in 16 seconds on our reference system, a six-core 2019 Mac mini, instead of 25 seconds for the BE1.6 configuration script. This is not a significant improvement for users, but it is for BE developers. The compile time for both is 32 seconds using make-j.
The C++ codebase was reduced by approximately 10,000 lines, or 17%, even accounting for the lines added for the new unit tests. The program is now approximately 46,000 lines of C++ and GNU flex code.
Dynamic analysis with unit tests and test coverage. As hinted previously, despite its widespread use, BE lacked a modern approach to testing. Specifically, the BE codebase was devoid of systematic unit tests. Instead, the program was occasionally run on test datasets during the development process, and the output was manually compared with that from previous runs: If the two outputs were substantially similar, the program was deemed to be not obviously broken.

For BE2, we decided to implement unit tests for all levels of the BE source code, including low-level data manipulation routines, forensic scanners, result-reporting code, option processing, and end-to-end tests. A review of the C++ unit test frameworks on Wikipedia led to Catch2,6 which has support for test scaffolds, implements a minimal CLI (command-line interface), can test for the presence (or absence) of thrown exceptions, and appears to be well supported and maintained.

We also enabled AddressSanitizer22 by default on the development system. (We enabled ThreadSanitizer23 and found several thread-sharing errors but encountered a false positive resulting from a conflict between one of its heuristics and our multithreading paradigm, preventing it from being enabled by default.)

We started with unit tests for the BE2 framework, generally writing them as the new interfaces were designed and implemented, combining the creation of each new test with related refactoring. The code coverage of the unit tests was tracked and systematically increased, then integrated with GitHub's "actions" system, so that the tests would run on every push. The popular CodeCov.io website displayed the code coverage results.

After all the new and refactored code had unit tests, the code coverage reports revealed which pieces of legacy code were not covered by the newly written unit tests. The target code coverage was 60%. In some cases, legacy code was covered by the new tests, because the new code called the old code. For about two-thirds of the legacy code, however, there was no test coverage, which we resolved by writing new tests from scratch. At first, this felt like a pointless compliance exercise—after all, BE had been in use for more than a decade, so we assumed that significant bugs, such as memory allocation errors and off-by-one errors, would not be present in the codebase. The act of writing the unit tests, however, forced us to clarify internal documentation, simplify internal implementations, and in some cases, eliminate legacy code that was no longer being used. We even found a few dormant bugs!

To improve encapsulation and provide for better code reuse, many functions were moved from scanners and the BE framework into the sbuf.

Removing functionality. In addition to removing experimental functionality, we improved performance of BE by disabling scanners that were computationally intensive but that rarely if ever extracted useful forensic data.

For example, we now disable the hibernation file (hiberfil) scanner (xpress decompression) by default, because we lack test vectors that could be used to demonstrate the correctness of our implementation, and because Windows may no longer be using the compression algorithm that BE1 implemented.

We also disabled (by default) scanning for 192-bit [[AES]] (Advanced Encryption Standard) keys in memory, because AES is rarely used in its 192-bit mode.

All the features disabled by default can be re-enabled with command-line options. We also removed key functionality that was not being used:

Internet search engines were used to see if some of the program's more obscure command-line options were being referenced in open source programs, scripts, or even blog entries that provide tutorials for using BE. Obscure options that were unused by the user community were eliminated.
To the best of our knowledge, no one (other than the original developer) ever used BE's shared library to let the program's scanner system be called from C++ or Python, so we removed it, although it could be re-implemented in the future.
The ability to load scanners as shared libraries at startup has not been updated for BE2, although this update is trivial and will be implemented if users request it.
Incompatible changes. Despite efforts to retain full compatibility between BE1 and BE2, a few minor incompatible changes were required in the interest of correctness and modernization:

BE feature files are UTF-8, but some of the information in them is binary and must be escaped. In BE1, non-Unicode characters were present and escaped in octal. In BE2, non-Unicode characters are escaped in hexadecimal.
A persistent problem is how UTF-16 features should be represented in the UTF-8 feature files. BE1 presented UTF-16 as octal-escaped values, which were hard to read. BE2 converts UTF-16 into UTF-8 in the second ("feature") column but leaves the features as (escaped) UTF-16 in the third ("context") column.
The ZIP scanner was not properly reporting the location of artifacts within ZIP-decoded data blocks; this was corrected (addressed later in this article).
We dropped support for MD5 (message-digest algorithm) in BE2, and now use SHA-1 (secure hash algorithm-1).2 (Sadly, the DF community has been slow to move to SHA-256 or SHA-3.)
Performance tuning. Despite the effort to eliminate all memory copies, an interim version of BE2 was dramatically slower than BE1.6. For example, scanning the 2009-domexusers15 (an NTFS file on a computer running Windows XP containing two user accounts) disk image on the reference Mac mini required approximately 10 minutes with BE1.6 but took 70 minutes with the BE2 development version.

BE has long had the ability to measure each scanner's contribution to runtime. Specifically, it keeps counters (in std::atomic<> variables) of how many times each scanner is called and how many nanoseconds it spends executing. These counters revealed that just three scanners (rar, net, and aes) were responsible for most of the time spent scanning.

Each of these scanners has a hand-coded loop that scans through the memory image, scanning for a magic number. The loop had been re-implemented and was now making a new sbuf for each location. We removed the loop and moved the scanning function into the sbuf class implementation itself, which eliminated the need to create and destroy an sbuf for each character.

We cannot stress enough that using the program to measure and improve its own performance turned out to be more effective than using off-the-shelf performance-monitoring tools.

Once these changes were made, the rar scanner was no longer the slowest. Now the slowest scanners were net, aes, and the flex-based email and accts (but not the other flex-based scanners, curiously enough).

We iterated. With micro-benchmarks and more testing, we were able to identify many other opportunities for speedup. We cannot stress enough that using the program to measure and improve its own performance turned out to be more effective than using off-the-shelf performance-monitoring tools.

Validation. The forensic tool community is increasingly turning its attention to developing specifications and tests for various aspects of a tool's intended operation. After all, "a program that has not been specified cannot be incorrect; it can only be surprising."25

We performed two kinds of validation on BE2: correctness and throughput. For correctness, BE2 needed to produce results that were as good as the results of BE1. For throughput, BE2 needed to be at least as fast as BE1.

Correctness. When differences occurred between the output of BE1 and BE2, some were cases in which BE2 was correct. In these cases, it appeared that the BE1 output had never been validated in detail. Most of these had to do with the location of recursively analyzed features in the feature file.

All DF tools require the ability to specify the location in evidence from which a feature, such as an email address, has been found. With BE, the ability to specify such locations is complicated by the fact that a byte-stream might need to be decoded, decompressed, or otherwise transformed. BE1 introduced the concept of a forensic path, which allows the specification of both a location and one or more transformations. For example, the BE forensic path 456536-ZIP-1255117 is read to mean that a feature is located 1,255,117 bytes into an inflated ZIP stream that is itself located 456,536 bytes from the beginning of the disk image.

During the development of unit tests, we discovered that some forensic path locations reported by BE1 failed to include six bytes of the ZIP header. This was corrected by adding new code so the sbuf class computes the offset, rather than hard coding the calculation into each scanner. This specific error was discovered while writing a unit test for the forensic path printer—the part of BE that reads a forensic path, performs the specified transformations, and performs a hexdump of the evidence. Although this code had been in use for more than 10 years in the BE, apparently it had never worked properly for the ZIP scanner, and none of BE's users had ever reported it not working. (The GZIP scanner reported forensic paths correctly.)

Many of the code paths in the BE1 code-base were painstakingly developed on specific test cases, but those cases had not been added to the code-base as unit tests. They were added in BE2, assuring that the tests would automatically run using GitHub's "Actions" continuous integration facilitation on every commit.

Throughput. Measuring the speed with which BE processes a disk image or other form of electronic evidence is straightforward. Explaining variations in speed is significantly harder. The time that BE spends processing evidence is highly dependent upon the contents: A disk image that contains many compressed archives will take longer to process because each compressed run of bytes will be decompressed and recursively reanalyzed. A disk that is filled with JPEGs will analyze quickly, but if carving mode 1 is enabled, each JPEG will need to be copied into a new file. If carving mode is set to 2 (the default), however, only the JPEGs that had to be decompressed or otherwise decoded—the JPEGs typically missed by other carving tools—will be copied.

BE also incorporates many techniques to discard data before applying the full recursive analysis. For example, duplicate data is not analyzed a second time, which has the side benefit of protecting the program from compression bombs. Likewise, pages that consist of a repeating n-gram (for example, @ABCABCABC…) will not be analyzed, unless a scanner indicates in its metadata that such analysis is desired.

Another factor in performance is the computer on which the program is run. The number of CPU cores, amount of RAM, speed of that RAM, and speed of the I/O system all impact throughput. All these factors interact with the evidence under examination: A disk image that has a lot of blank and repeated sectors will benefit more from a faster I/O system, while a disk image with a lot of complex data structures will benefit more from additional cores.

Therefore, throughput and benchmark results are best reported using evidence that is ecologically valid,1 such as an actual disk image. Although such media are commonly used in software development and internal benchmarking, they tend not to be publicly released for privacy reasons.

The accompanying table shows the performance of BE1.6 and BE2 with three reference disk images from the Digital Corpora collection, running on three different reference computers. The disk images are nps-2009-ubnist1, a 2.1GB disk image of a bootable USB drive running Ubuntu Linux; nps-2009-domexusers, a 42GB disk image of a Microsoft Windows system that was used by several individuals in a lab; and nps-2011-2tb, a 2.0TB disk image containing the entire GovDocs1 corpus and several other of the Digital Corpora reference disk images. These images were made at the Naval Postgraduate School between 2009–2011 and are hosted on the Digital Corpora website (digitalcorpora.org).

ut1.jpg
Table. Clock time comparison of running BE1.6 and BE2 on a variety of hardware and software configurations.

Times for the nps-2009-ubnist1 and nps-2009-domexusers are averages of three runs. The nps-2009-ubnist1 and nps-2009-domexusers read and write to the system SSD (solid-state drive), while the nps-2013-2tb reads from the system SSD and writes to an external USB3 hard drive because of storage considerations. BE1.6 speeds are reported for runs with the standard 30 default scanners enabled: accts, aes, base64, elf, email, evtx, exif, find, gps, gzip, hiberfile, httplogs, json, kml, msx-ml, net, ntfsindx, ntfslogfile, ntfsmft, ntfsusn, pdf, rar, sqlite, utmp, vcard, windirs, winlnk, winpe, winprefetch, and zip. BE2 speeds are for runs with the standard 29 scanners enabled (hiberfil is disabled) and with AES192 key searching disabled, and with the BE1.6 configuration that adds hiberfil and AES192 key searching. The Apple M1 Pro 10 core processor has eight performance cores and two efficiency cores. Throughput is normalized to the speed of BE1.6 on the same hardware with the same disk image; a throughput of 200% means that the disk image will be analyzed in half the time.

Performance is reported using three Apple Macintosh computers. Both BE1.6 and BE2 were compiled on a computer on which the benchmark was run with the current LLVM compiler provided by Apple. All compilation was done with optimization level -O3, with both AddressSanitizer and ThreadSanitizer disabled. We report BE1.6 and BE2 with the default analysis. In this configuration, 30 scanners are enabled for BE1.6, but BE2 disables hiberfil and AES192 key searching. For this reason, BE2 is also reported with the BE1.6 configuration (the rightmost two columns). As can be seen in table, BE2 is faster than BE1.6 in nearly every case, although the speedup is more pronounced on modern hardware with more cores.

Recommendations and Future Work
This multiyear exercise shows the value of updating tools to current software engineering practices, even when they appear to be working and bug-free. We recommend a scrub of all modern digital forensic tools, as rewriting these tools will likely make them faster and more reliable.

Reading Stroustrup's book was time-consuming preparation for this project, but well worth the investment. A similar benefit was derived from reading the entire Python reference manual prior to embarking on a large-scale Python project. Detailed reading of all developer documentation for implementation languages and tools is a good idea. Organizations investing in DF research and tools should also be prepared to invest for the long term, to provide for maintenance, adaptation, and growth of promising tools, as well as focused attention for developers.

The improvement in code quality that resulted from the pursuit of 60% unit-test code coverage was stunning. The power of AddressSanitizer in finding a wide variety of bugs was also surprising. Test-driven development3 and test-driven refactoring11 should be adopted as primary tools, and Address-Sanitizer should always be enabled during the development process.

The speed of C++ compared with Python is a clear incentive to use this language for speed-critical applications. Given the lack of C++ programmers in the DF community, however, it is clear BE requires an interface to allow Python scanners to be called. Because Python is not thread-safe, a separate Python interpreter is required for each analysis thread. C++ should be used with well-designed classes to provide memory safety, and Python-based APIs should be provided to access their functionality.

We achieved a 61% code coverage for the BE2 framework but only 47% for the remainder of the BE2 codebase (excluding the framework). Clearly there is still room for improvement.

Finally, the increased use of file system-level compression and encryption, combined with the use of the TRIM command on SSDs, means that the bulk data analysis of raw storage devices is likely to yield less data in the future than a systematic extraction of bulk data from resident files. That is, running BE2 with the -r (recursive) option on a mounted file system may one day yield more useful information than running it on the raw device. Ideally, it would be possible to run BE2 on a file system, keep track of the sectors that were scanned, and then process the remaining sectors raw. Another approach would be to perform two passes: one on the mounted files and another on the raw device. Evaluation of these strategies is left as future work.

==References==
<references />
===Other Material===
1. Andrade, C. Internal, external, and ecological validity in research design, conduct, and evaluation. Indian J. Psychological Medicine 40, 5 (2018), 498–499; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6149308/.

2. Barker, R., Roginsky, A. Transitioning the use of cryptographic algorithms and key lengths. National Institute of Standards and Technology, Special Publication 800–131A, rev. 2, 2019; https://bit.ly/3KqdgKH.

3. Beck, K. Test-Driven Development: By Example. Addison-Wesley Professional, 2002.

4. BitCurator; https://bitcurator.net/.

5. Brooks, F.P. The Mythical Man-Month—Essays on Software Engineering. Addison-Wesley, 1975.

6. Catch2. GitHub; https://github.com/catchorg/Catch2.

7. Cellebrite. Blackbag technology software user license agreements; https://cellebrite.com/en/blackbag-agreements/.

8. Donnelly, C., Stallman, R.M. Bison—the Yacc-compatible parser generator. Free Software Foundation; https://www.gnu.org/software/bison/manual/

9. cxxopts. GitHub; https://github.com/jarro2783/cxxopts.

10. Flex—fast lexical analyzer generator. GNU software package; https://github.com/westes/flex.

11. Fowler, M. Refactoring: Improving the Design of Existing Code, Second Edition. Addison-Wesley Professional, 2018.

13. Garfinkel, S.L. Digital media triage with bulk data analysis and bulk _ extractor. Computers and Security 32, C (2013), 56–72; https://dl.acm.org/doi/10.5555/2748150.2748581.

14. Garfinkel, S. SBook: Simson Garfinkel's Address Book, Version 2.0. Simson Garfinkel and Assoc., 1992; https://simson.net/ref/1992/SBook20.pdf.

15. Garfinkel, S.L., Farrell, P., Roussev, V., Dinolt, G. Bringing science to digital forensics with standardized forensic corpora. In Proceedings of the 9th Annual Digital Forensic Research Workshop 6, 2009 (supplement); https://bit.ly/3nZXPBj.

16. Garfinkel, S. Digital media triage with bulk data analysis and bulk _ extractor. Computers and Security 32 (2013) 56–72.

17. Garfinkel, S., Shelat, A. Remembrance of data passed. IEEE Security and Privacy 1, 1 (2003), 17–27; https://dl.acm.org/doi/abs/10.1109/MSECP.2003.1176992.

18. Liu, J., Moore, R.T. An overview of the NSA's declassified intelligence oversight board reports. Lawfare, 2015; http://bit.ly/41dlgp1.

19. Marlinspike, M. Exploiting vulnerabilities in cellebrite UFED and physical analyzer from an app's perspective. Signal, 2021; https://signal.org/blog/cellebrite-vulnerabilities/.

20. National Institute of Justice. Forensic examination of digital evidence: a guide for law enforcement, 2004; https://www.ojp.gov/pdffiles1/nij/199408.pdf.

21. Office of Deputy Assistant Secretary of Defense, Systems Engineering. Value engineering: a guidebook of best practices and tools, 2011; https://bit.ly/3Uw8Syo

22. Serebryany, K., Bruening, D., Potapenko, A., Vyukov, D. AddressSanitizer: a fast address sanity checker. In Proceedings of the 2012 Usenix Annual Technical Conference 28; https://dl.acm.org/doi/10.5555/2342821.2342849.

23. Serebryany, K., Iskhodzhanov, T. ThreadSanitizer: data race detection in practice. In Proceedings of the 2009 Workshop on Binary Instrumentation and Applications, 62–71; https://dl.acm.org/doi/10.1145/1791194.1791203.

24. Stroustrup, B. The C++ Programming Language, Fourth Edition. Addison-Wesley, 2013; https://www.stroustrup.com/4th.html.

25. Young, W.D., Boebert, W., Kain, R. Proving a computer system secure. The Scientific Honeyweller 6, 2 (1985), 18–27.

Transformer

2026-03-08T04:34:10Z

Tom: /* References */

==Meme==
A [[Transformer]] in AI is a type of neural network architecture that revolutionized how machines process sequential data—like text, audio, or even DNA sequences.
==Context==
summary of the 2017 paper “[[Attention]] Is All You Need” by Vaswani et al., which introduced the [[Transformer]] architecture—a paradigm shift in the approach sequence modeling in deep learning:

The paper proposes a novel architecture called the Transformer, which relies entirely on attention mechanisms—no recurrence (RNNs) or convolutions (CNNs).

This design enables parallel processing, faster training, and better handling of long-range dependencies in sequences2.

=== Key Components===
Self-Attention: Each token in a sequence attends to every other token, allowing the model to weigh relationships dynamically.

Multi-Head Attention: Multiple attention layers run in parallel, capturing diverse aspects of token relationships.

Positional Encoding: Since the model lacks recurrence, positional encodings inject order information into the input sequence.

Encoder-Decoder Structure: Both parts use stacked layers of attention and feed-forward networks to process and generate sequences.

⚙️ Technical Advantages
Scalability: Fully parallelizable, making it faster to train than RNNs or CNNs.

Performance: Achieved state-of-the-art results on machine translation tasks (e.g., WMT 2014 English-German and English-French).

Generalization: Successfully applied to other NLP tasks like parsing, showing broad applicability.

===Impact===
Sparked a wave of innovation leading to models like BERT, GPT, T5, and PaLM.

Became one of the most cited papers in AI, reshaping the landscape of natural language processing and beyond.

==Functionality==
Grounded in a few key properties the literature consistently emphasizes: **self‑attention**, **parallel sequence processing**, and **the quadratic cost of attention with context length**.

## What a Transformer architecture **can do**
===Learn rich *contextual representations* of sequences===

Transformers use **self‑attention** to let each token (word/patch/element) “look at” other tokens and weigh what matters for meaning. This makes them excellent at capturing **relationships** and **dependencies** across a sequence (e.g., which earlier words constrain later words).

**Practical outcome:** strong performance on tasks where meaning depends on context:

* translation, summarization, Q\&A, text generation
* classification and extraction
* coding assistance (because code is also a sequence with long-range dependencies)

(These are exactly the kinds of capabilities described in accessible transformer overviews.)

===Train efficiently at scale (parallelizable sequence processing)===

Unlike RNNs/LSTMs (which process tokens sequentially), Transformers can process many tokens **in parallel** during training because attention does not require strict step-by-step recurrence. This is one reason they scaled to very large models and datasets.

**Practical outcome:** they are the workhorse behind “foundation models” that learn broad language/vision patterns from massive corpora.

===Generalize patterns and do “soft” reasoning within the learned distribution===

Transformers can often generalize from examples and infer implications by associating concepts—sometimes impressively. But that same mechanism can also create hallucinations when the association is spurious rather than causal. This duality is studied explicitly as “out-of-context reasoning” in Transformers.

**Practical outcome:** they can do plausible multi-step explanations and pattern-based reasoning, especially when the domain is well represented in training data—but the same machinery can produce confident-sounding errors.

===Serve as a universal backbone (text, vision, multimodal)===

The core idea (attention over a sequence) has been adapted beyond language—e.g., “Vision Transformers” and multimodal models. High-level transformer explainers emphasize how the architecture generalizes to many sequence-like data forms.

===What a Transformer architecture **cannot do (by itself)**===

====It cannot guarantee truth, verification, or correctness====

A Transformer is fundamentally a **sequence model** that maps an input token sequence to an output token sequence. It does not inherently include a “truth-checking” module. This becomes especially relevant when tasks require verification beyond what the model can reliably compute or validate internally.

A recent complexity-based critique argues that transformers have inherent limits in performing and *verifying* tasks whose computational demands exceed what the model’s inference computation can cover, linking this to hallucinations and unreliability in high-stakes verification.

**Practical consequence:**
Transformers alone are not trustworthy as final arbiters for:

* factual verification
* high-precision calculations
* rigorous proof checking
* auditing their own outputs

(You can *reduce* error with retrieval/tools, but the architecture alone doesn’t guarantee correctness.)

====It cannot naturally handle arbitrarily long context (without tradeoffs)====

Vanilla self‑attention has a well-known **quadratic scaling** with input length, often summarized as a major limitation. One AAAI paper states directly that a “major limitation” for broader problem scope is the quadratic scaling of compute with input size, and proposes recurrent memory augmentation to address it.

**Practical consequence:**
Even with extended context windows in some systems, long documents and long-running tasks can still suffer from:

* high compute cost
* degraded retrieval across long spans
* “lost in the middle” effects (in practice)
* the need for chunking / memory / retrieval augmentation

====It cannot reliably plan and execute multi-step real-world actions on its own====

Transformers can *describe* plans and *simulate* steps in text, but robust real-world planning typically requires:

* persistent memory/state
* explicit world models
* tool use and error correction loops
* grounded feedback from the environment

The complexity-limit argument explicitly flags difficulties for “agentic” uses beyond certain complexity and warns against relying on LLMs for critical verification in autonomous settings.

**Practical consequence:**
For reliable agents, organizations typically combine Transformers with:

* external tools (calculators, search, databases)
* planners/schedulers
* monitoring and guardrails
* logging and evaluation harnesses

====It cannot “ground” symbols in the physical world without extra machinery====

A pure Transformer processes tokens; it doesn’t inherently connect those tokens to real-world referents (perception, embodied interaction, sensors, etc.). That’s why practical systems use multimodality, retrieval, tools, or embodied agents when grounding matters.

(We discussed earlier the “messy real world” problem—Transformers alone don’t solve that; they operate on representations.)

## A helpful rule of thumb

**Transformers are amazing at:**

* mapping patterns in sequences to sequences
* compressing massive experience into predictive representations
* generating coherent, context-sensitive outputs

**Transformers are weak at (unless augmented):**

* guaranteed correctness and verification
* long-horizon memory and very long context (quadratic attention cost)
* grounded, reliable real-world agency under uncertainty

## Quick “capability / limitation” checklist (practical)

**Good fit for Transformers**

* Drafting, summarizing, transforming text
* Extracting structure from language (entities, intents, patterns)
* Pattern generalization from examples (with caveats about hallucinations)

**Not a good fit *without augmentation***

* “Be correct all the time” tasks (math proofs, audits, medical/legal verification)
* Very long-document consistency without memory/retrieval augmentation
* Autonomous operation where mistakes are costly and feedback is sparse

====Small bonus (grounded to your workspace)====

I noticed an unread email from [ACM Bulletin](https://outlook.live.com/people/0/ACM+Bulletin?EntityRepresentationId=1d604da3-15c1-46ac-8305-052bdf857626) mentioning “transformer with pretraining” as one of the deep learning advances shaping multimedia research. If you want, I can summarize that email’s key point in one paragraph and connect it to why Transformers became dominant.

==Solutions==
I’ll structure this as **problem → why Transformers struggle → what to add around them**, because the key lesson is that *you don’t “fix” Transformers by scaling alone; you fix them by system design*.

How to Address Transformer Limitations in Your Project
===Limitation: **Hallucinations & lack of truth guarantees**===

### Why this happens

Transformers are **sequence predictors**, not truth verifiers. They generate the most likely next token given context, even when the task implicitly requires verification. Recent work explicitly links hallucinations to **architectural and computational complexity limits**, not just data quality.

### What to do instead

**Externalize truth.**

Add components that *the Transformer must defer to*:

* Retrieval systems (databases, search indexes, knowledge graphs)
* Deterministic tools (calculators, compilers, validators)
* Rule‑based or symbolic checks for high‑stakes constraints

**Pattern to use:**

> *Transformer proposes → external system verifies → Transformer explains*

This design directly reflects warnings in the literature against using LLMs as standalone verifiers.

===Limitation: **Long‑context fragility and memory decay**===

### Why this happens

Self‑attention scales quadratically with input length, which limits how reliably a Transformer can reason over very long sequences. Even when context windows are expanded, performance degrades across long spans.

### What to do instead

**Chunk, retrieve, and summarize—don’t rely on raw context.**

Effective patterns include:

* Retrieval‑Augmented Generation (RAG)
* Hierarchical summaries (rolling or layered)
* Explicit memory stores (vector databases, logs)

Research exploring recurrent memory augmentation exists precisely because vanilla Transformers struggle here.

=== Limitation: **Weak planning and unreliable multi‑step reasoning**===

### Why this happens

Transformers can *describe* plans but are not inherently designed to **execute and monitor long‑horizon plans**, especially under uncertainty. Complexity‑based analyses show clear limits for agentic tasks beyond certain thresholds.

### What to do instead

**Separate planning from language generation.**

Use:

* Explicit planners or workflows
* Step‑wise execution with checkpoints
* Feedback loops after each action

**Pattern to use:**

> *Planner decides steps → Transformer handles language & interpretation → environment feeds back results*

This avoids overloading the Transformer with responsibilities it was never designed for.

===Limitation: **Lack of grounding in the real world**===

### Why this happens

Transformers operate on **tokens**, not on physical or social reality. This is a classic instance of the **symbol grounding problem**: symbols do not carry intrinsic meaning without interaction with the world.

### What to do instead

**Ground via interfaces, not definitions.**

Options include:

* Multimodal inputs (images, sensors, structured data)
* Tool calls that affect real systems
* Human‑in‑the‑loop checkpoints

Do not assume better prompts or labels alone will solve grounding; the literature is clear that meaning emerges from interaction, not annotation.

=== Limitation: **Opacity and interpretability**===

### Why this happens

Transformer internals are high‑dimensional and learned implicitly. As noted in research discussions, even their creators often cannot fully explain *why* a given output appears.

### What to do instead

**Design for observability at the system level.**

You can’t fully “open the black box,” but you can:

* Log inputs, outputs, and tool calls
* Track confidence signals and failure modes
* Run adversarial tests and red‑teaming

The Quanta Magazine discussion emphasizes that disagreement about AI capabilities persists partly because internal mechanisms are poorly understood.

=== A simple architecture rule (that actually works)===

> **Use Transformers for language, pattern recognition, and synthesis—not for authority, memory, or control.**

In robust systems:

* Transformers are *one component*
* Verification, memory, planning, and safety live elsewhere
* Failure is expected and handled, not denied

This design philosophy aligns with both the academic critiques and real‑world deployment lessons surfaced across recent literature.

===A quick diagnostic checklist for your project===

Ask yourself:

* **What happens if the model is confidently wrong?**
→ Add verification.
* **What happens if context is too long or missing?**
→ Add retrieval/memory.
* **Does this require planning or action?**
→ Add a planner or workflow engine.
* **Does meaning depend on the real world?**
→ Add grounding interfaces.
* **Is this safety‑critical?**
→ Never rely on the Transformer alone.

===Bottom line===

Transformers are extraordinarily powerful **language and pattern engines**—but they are not:

* truth engines
* planners
* grounded agents
* or safety guarantees

You address their limitations not by asking them to be smarter, but by **designing systems that assume they will fail gracefully**.

If you want, tell me:

* what domain your project is in (research, product, safety‑critical, exploratory),
* whether it’s interactive or batch,
* and where failure would be most costly,

==References==

[[Category: Artificial Intelligence]]

Transformer

2026-03-08T04:28:43Z

Tom: /* References */

Dialectic Harmony

2026-03-08T04:20:20Z

Tom: /* What “mystical” means here (and what it does not) */

==Meme==
[[Dialectic Harmony]] is a philosophical and aesthetic idea that sees harmony not as the absence of conflict, but as the dynamic balance that emerges from opposites in tension. Instead of imagining harmony as smooth uniformity, it treats contradiction as the very generator of order and beauty.

==Context==
[[Dialectic Harmony]] is a philosophical and aesthetic idea that sees **harmony not as the absence of conflict, but as the dynamic balance that emerges from opposites in tension**. Instead of imagining harmony as smooth uniformity, it treats contradiction as the very generator of order and beauty.

* Roots of the Concept
** Heraclitus (5th c. BCE, Greece)
*** Famously said: *“The hidden harmony is better than the obvious.”*
*** For him, harmony arises from the **strife of opposites** — day and night, war and peace, life and death.
*** This is sometimes called **dialectical monism**: unity through contradiction.
** Chinese Aesthetics (Confucian & Daoist traditions)
*** Harmony (*he*, 和) is not uniformity but the **balanced interplay of differences**.
*** Think yin and yang: opposites that generate wholeness through their interaction.
** Modern Comparative Philosophy
*** Scholars like Tak‑lap Yeung describe two types of harmony:
*** *Dialectic harmony**: harmony through conciliation of opposites.
*** *Ambiguous harmony**: harmony through coexistence of diversity without resolution.
* What Dialectic Harmony Means
**Not peace through sameness**, but peace through balance.
**Not erasing contradictions**, but letting them interact productively.
**Not static equilibrium**, but a living, dynamic process.
* Examples
**Music**: Counterpoint in Bach — independent voices in tension, yet forming a higher unity.
**Politics**: A pluralist democracy — conflicting interests negotiated into a workable order.
**Digital Commons**: Privacy vs. transparency, human vs. agent control — contradictions institutionalized as engines of trust.
*Takeaway
** Dialectic harmony is the art of **making opposites resonate**. It’s a philosophy of becoming, where contradiction is not a flaw but the **source of deeper unity**.

==Scientific vs Mystical==
Thesis
Clausewitz’s coup d’œil rests on a mystical epistemology of insight; Boyd’s OODA loop rests on a scientific epistemology of adaptation.
They may address similar problems—decision‑making under uncertainty—but they do so from opposite intellectual traditions.

===What “mystical” means here (and what it does not)===
Calling coup d’œil “mystical” does not mean:
supernatural
irrational
anti‑intellectual
It means:
non‑procedural
non‑replicable
non‑falsifiable
grounded in inner vision rather than external testing
By contrast, “scientific” means:
model‑based
iterative
falsifiable
improvable through feedback
This distinction aligns cleanly with how Clausewitz and Boyd each frame decision authority.

===Coup d’œil: insight without method (mystical)===
<pre>
a. Immediate apprehension, not process
Clausewitz defines coup d’œil as the commander’s ability to see the truth of a situation at a glance, without calculation or analysis. It is an instantaneous synthesis of complexity that “the mind would ordinarily miss” without prolonged study.
There is no algorithm:
no steps
no cycle
no feedback loop
The knowledge arrives whole.
This places coup d’œil squarely in the tradition of gnosis = knowing by direct apprehension rather than by inference.
b. Inseparable from “genius”
Clausewitz embeds coup d’œil inside his concept of military genius, alongside courage, character, and strength of will.
Crucially:
Genius is rare
Genius is unevenly distributed
Genius cannot be manufactured through training alone
That makes coup d’œil elitist and mystical by design: it is a gift some possess and others do not. Clausewitz explicitly rejects the idea that war can be reduced to rules that would make great commanders common.
This mirrors older traditions:
Aristotle’s phronesis
Romantic intuition
even theological notions of “grace”
c. Immunity to falsification
If a commander with coup d’œil succeeds, it proves the insight was real.
If they fail, Clausewitz attributes the failure to friction, chance, or political interference—not to a flaw in the faculty itself.
That makes coup d’œil non‑falsifiable.
A claim that cannot be tested, replicated, or disproven may be profound = but it is not scientific.
</pre>
=== OODA loop: decision as a testable system (scientific)===
<pre>
a. Explicit model, explicit stages
Boyd’s OODA loop is a formal model:
Observe
Orient
Decide
Act
Even in its more complex, non‑linear form, it remains a system description of how decisions occur in competitive environments.
A model can be:
trained
simulated
improved
broken
That alone places OODA firmly in the scientific tradition.
b. Orientation as hypothesis formation
Boyd’s key contribution is not speed but Orientation—the mental models through which observations are interpreted. Orientation is continuously reshaped by:
feedback
error
surprise
failure.
This is textbook scientific epistemology:
Observe reality
Apply a model
Act
Update the model based on results
Boyd explicitly draws on:
cybernetics
evolutionary theory
systems thinking
Where Clausewitz asks for vision, Boyd asks for learning.
c. Falsifiability and improvement
An OODA loop can fail:
bad observation
distorted orientation
slow decision
ineffective action
When it fails, the failure is diagnosable.
That is the decisive difference.
A unit, organization, or individual can:
shorten loops
improve sensing
diversify mental models
redesign feedback channels
OODA does not require genius. It assumes fallibility and designs around it.
That is the scientific posture.
</pre>
===Dialectics vs. dynamics===
<pre>
Clausewitz: dialectical truth
Clausewitz’s worldview is Hegelian:
truth emerges from contradiction
war resists equilibrium
understanding is synthetic, not analytic
Coup d’œil is the mind’s ability to stand above contradiction without resolving it.
This is philosophy, not science.
Boyd: dynamical systems
Boyd’s worldview is Darwinian and cybernetic:
systems adapt or collapse
faster learning beats brute force
coherence is temporary
Victory comes not from transcendent insight but from continuous model destruction and renewal.
This is applied science.
5. Why this distinction matters
The difference is not academic—it predicts how each scales.
Dimension
Coup d’œil
OODA Loop
Authority
Individual
System
Replicability
No
Yes
Teachability
Limited
High
Debuggability
None
Strong
Suitability for AI
Poor
Excellent

Modern militaries, businesses, and AI systems overwhelmingly adopt Boyd, not Clausewitz, because:
machines cannot have mystical insight
but they can run loops
update models
and learn from error
6. Final claim (clean and sharp)
Coup d’œil is mystical because it locates truth inside the commander; OODA is scientific because it locates truth in the interaction between model and reality.
Clausewitz trusts vision.
Boyd trusts adaptation.
That is the fault line between pre‑modern strategic thought and modern systems thinking.
</pre>

==AI overtakes the web==

==References==
* See Wiki [[OODA]] for John Boyd's take on [[Dialectic Harmony]]

[[Category: Philosophy]]

Dialectic Harmony

2026-03-08T04:15:19Z

Tom: /* OODA loop: decision as a testable system (scientific) */

==Meme==
[[Dialectic Harmony]] is a philosophical and aesthetic idea that sees harmony not as the absence of conflict, but as the dynamic balance that emerges from opposites in tension. Instead of imagining harmony as smooth uniformity, it treats contradiction as the very generator of order and beauty.

==Context==
[[Dialectic Harmony]] is a philosophical and aesthetic idea that sees **harmony not as the absence of conflict, but as the dynamic balance that emerges from opposites in tension**. Instead of imagining harmony as smooth uniformity, it treats contradiction as the very generator of order and beauty.

* Roots of the Concept
** Heraclitus (5th c. BCE, Greece)
*** Famously said: *“The hidden harmony is better than the obvious.”*
*** For him, harmony arises from the **strife of opposites** — day and night, war and peace, life and death.
*** This is sometimes called **dialectical monism**: unity through contradiction.
** Chinese Aesthetics (Confucian & Daoist traditions)
*** Harmony (*he*, 和) is not uniformity but the **balanced interplay of differences**.
*** Think yin and yang: opposites that generate wholeness through their interaction.
** Modern Comparative Philosophy
*** Scholars like Tak‑lap Yeung describe two types of harmony:
*** *Dialectic harmony**: harmony through conciliation of opposites.
*** *Ambiguous harmony**: harmony through coexistence of diversity without resolution.
* What Dialectic Harmony Means
**Not peace through sameness**, but peace through balance.
**Not erasing contradictions**, but letting them interact productively.
**Not static equilibrium**, but a living, dynamic process.
* Examples
**Music**: Counterpoint in Bach — independent voices in tension, yet forming a higher unity.
**Politics**: A pluralist democracy — conflicting interests negotiated into a workable order.
**Digital Commons**: Privacy vs. transparency, human vs. agent control — contradictions institutionalized as engines of trust.
*Takeaway
** Dialectic harmony is the art of **making opposites resonate**. It’s a philosophy of becoming, where contradiction is not a flaw but the **source of deeper unity**.

==Scientific vs Mystical==
Thesis
Clausewitz’s coup d’œil rests on a mystical epistemology of insight; Boyd’s OODA loop rests on a scientific epistemology of adaptation.
They may address similar problems—decision‑making under uncertainty—but they do so from opposite intellectual traditions.

===What “mystical” means here (and what it does not)===
<pre>
Calling coup d’œil “mystical” does not mean:
supernatural
irrational
anti‑intellectual
It means:
non‑procedural
non‑replicable
non‑falsifiable
grounded in inner vision rather than external testing
By contrast, “scientific” means:
model‑based
iterative
falsifiable
improvable through feedback
This distinction aligns cleanly with how Clausewitz and Boyd each frame decision authority.
</pre>
===Coup d’œil: insight without method (mystical)===
<pre>
a. Immediate apprehension, not process
Clausewitz defines coup d’œil as the commander’s ability to see the truth of a situation at a glance, without calculation or analysis. It is an instantaneous synthesis of complexity that “the mind would ordinarily miss” without prolonged study.
There is no algorithm:
no steps
no cycle
no feedback loop
The knowledge arrives whole.
This places coup d’œil squarely in the tradition of gnosis = knowing by direct apprehension rather than by inference.
b. Inseparable from “genius”
Clausewitz embeds coup d’œil inside his concept of military genius, alongside courage, character, and strength of will.
Crucially:
Genius is rare
Genius is unevenly distributed
Genius cannot be manufactured through training alone
That makes coup d’œil elitist and mystical by design: it is a gift some possess and others do not. Clausewitz explicitly rejects the idea that war can be reduced to rules that would make great commanders common.
This mirrors older traditions:
Aristotle’s phronesis
Romantic intuition
even theological notions of “grace”
c. Immunity to falsification
If a commander with coup d’œil succeeds, it proves the insight was real.
If they fail, Clausewitz attributes the failure to friction, chance, or political interference—not to a flaw in the faculty itself.
That makes coup d’œil non‑falsifiable.
A claim that cannot be tested, replicated, or disproven may be profound = but it is not scientific.
</pre>
=== OODA loop: decision as a testable system (scientific)===
<pre>
a. Explicit model, explicit stages
Boyd’s OODA loop is a formal model:
Observe
Orient
Decide
Act
Even in its more complex, non‑linear form, it remains a system description of how decisions occur in competitive environments.
A model can be:
trained
simulated
improved
broken
That alone places OODA firmly in the scientific tradition.
b. Orientation as hypothesis formation
Boyd’s key contribution is not speed but Orientation—the mental models through which observations are interpreted. Orientation is continuously reshaped by:
feedback
error
surprise
failure.
This is textbook scientific epistemology:
Observe reality
Apply a model
Act
Update the model based on results
Boyd explicitly draws on:
cybernetics
evolutionary theory
systems thinking
Where Clausewitz asks for vision, Boyd asks for learning.
c. Falsifiability and improvement
An OODA loop can fail:
bad observation
distorted orientation
slow decision
ineffective action
When it fails, the failure is diagnosable.
That is the decisive difference.
A unit, organization, or individual can:
shorten loops
improve sensing
diversify mental models
redesign feedback channels
OODA does not require genius. It assumes fallibility and designs around it.
That is the scientific posture.
</pre>
===Dialectics vs. dynamics===
<pre>
Clausewitz: dialectical truth
Clausewitz’s worldview is Hegelian:
truth emerges from contradiction
war resists equilibrium
understanding is synthetic, not analytic
Coup d’œil is the mind’s ability to stand above contradiction without resolving it.
This is philosophy, not science.
Boyd: dynamical systems
Boyd’s worldview is Darwinian and cybernetic:
systems adapt or collapse
faster learning beats brute force
coherence is temporary
Victory comes not from transcendent insight but from continuous model destruction and renewal.
This is applied science.
5. Why this distinction matters
The difference is not academic—it predicts how each scales.
Dimension
Coup d’œil
OODA Loop
Authority
Individual
System
Replicability
No
Yes
Teachability
Limited
High
Debuggability
None
Strong
Suitability for AI
Poor
Excellent

Modern militaries, businesses, and AI systems overwhelmingly adopt Boyd, not Clausewitz, because:
machines cannot have mystical insight
but they can run loops
update models
and learn from error
6. Final claim (clean and sharp)
Coup d’œil is mystical because it locates truth inside the commander; OODA is scientific because it locates truth in the interaction between model and reality.
Clausewitz trusts vision.
Boyd trusts adaptation.
That is the fault line between pre‑modern strategic thought and modern systems thinking.
</pre>

==AI overtakes the web==

==References==
* See Wiki [[OODA]] for John Boyd's take on [[Dialectic Harmony]]

[[Category: Philosophy]]

Dialectic Harmony

2026-03-08T00:10:02Z

Tom: /* AI overtakes the web */

==Meme==
[[Dialectic Harmony]] is a philosophical and aesthetic idea that sees harmony not as the absence of conflict, but as the dynamic balance that emerges from opposites in tension. Instead of imagining harmony as smooth uniformity, it treats contradiction as the very generator of order and beauty.

==Context==
[[Dialectic Harmony]] is a philosophical and aesthetic idea that sees **harmony not as the absence of conflict, but as the dynamic balance that emerges from opposites in tension**. Instead of imagining harmony as smooth uniformity, it treats contradiction as the very generator of order and beauty.

* Roots of the Concept
** Heraclitus (5th c. BCE, Greece)
*** Famously said: *“The hidden harmony is better than the obvious.”*
*** For him, harmony arises from the **strife of opposites** — day and night, war and peace, life and death.
*** This is sometimes called **dialectical monism**: unity through contradiction.
** Chinese Aesthetics (Confucian & Daoist traditions)
*** Harmony (*he*, 和) is not uniformity but the **balanced interplay of differences**.
*** Think yin and yang: opposites that generate wholeness through their interaction.
** Modern Comparative Philosophy
*** Scholars like Tak‑lap Yeung describe two types of harmony:
*** *Dialectic harmony**: harmony through conciliation of opposites.
*** *Ambiguous harmony**: harmony through coexistence of diversity without resolution.
* What Dialectic Harmony Means
**Not peace through sameness**, but peace through balance.
**Not erasing contradictions**, but letting them interact productively.
**Not static equilibrium**, but a living, dynamic process.
* Examples
**Music**: Counterpoint in Bach — independent voices in tension, yet forming a higher unity.
**Politics**: A pluralist democracy — conflicting interests negotiated into a workable order.
**Digital Commons**: Privacy vs. transparency, human vs. agent control — contradictions institutionalized as engines of trust.
*Takeaway
** Dialectic harmony is the art of **making opposites resonate**. It’s a philosophy of becoming, where contradiction is not a flaw but the **source of deeper unity**.

==Scientific vs Mystical==
Thesis
Clausewitz’s coup d’œil rests on a mystical epistemology of insight; Boyd’s OODA loop rests on a scientific epistemology of adaptation.
They may address similar problems—decision‑making under uncertainty—but they do so from opposite intellectual traditions.

===What “mystical” means here (and what it does not)===
<pre>
Calling coup d’œil “mystical” does not mean:
supernatural
irrational
anti‑intellectual
It means:
non‑procedural
non‑replicable
non‑falsifiable
grounded in inner vision rather than external testing
By contrast, “scientific” means:
model‑based
iterative
falsifiable
improvable through feedback
This distinction aligns cleanly with how Clausewitz and Boyd each frame decision authority.
</pre>
===Coup d’œil: insight without method (mystical)===
<pre>
a. Immediate apprehension, not process
Clausewitz defines coup d’œil as the commander’s ability to see the truth of a situation at a glance, without calculation or analysis. It is an instantaneous synthesis of complexity that “the mind would ordinarily miss” without prolonged study.
There is no algorithm:
no steps
no cycle
no feedback loop
The knowledge arrives whole.
This places coup d’œil squarely in the tradition of gnosis = knowing by direct apprehension rather than by inference.
b. Inseparable from “genius”
Clausewitz embeds coup d’œil inside his concept of military genius, alongside courage, character, and strength of will.
Crucially:
Genius is rare
Genius is unevenly distributed
Genius cannot be manufactured through training alone
That makes coup d’œil elitist and mystical by design: it is a gift some possess and others do not. Clausewitz explicitly rejects the idea that war can be reduced to rules that would make great commanders common.
This mirrors older traditions:
Aristotle’s phronesis
Romantic intuition
even theological notions of “grace”
c. Immunity to falsification
If a commander with coup d’œil succeeds, it proves the insight was real.
If they fail, Clausewitz attributes the failure to friction, chance, or political interference—not to a flaw in the faculty itself.
That makes coup d’œil non‑falsifiable.
A claim that cannot be tested, replicated, or disproven may be profound = but it is not scientific.
</pre>
=== OODA loop: decision as a testable system (scientific)===
<pre>
a. Explicit model, explicit stages
Boyd’s OODA loop is a formal model:
Observe
Orient
Decide
Act
Even in its more complex, non‑linear form, it remains a system description of how decisions occur in competitive environments.
A model can be:
trained
simulated
improved
broken
That alone places OODA firmly in the scientific tradition.
b. Orientation as hypothesis formation
Boyd’s key contribution is not speed but Orientation—the mental models through which observations are interpreted. Orientation is continuously reshaped by:
feedback
error
surprise
failure.
This is textbook scientific epistemology:
Observe reality
Apply a model
Act
Update the model based on results
Boyd explicitly draws on:
cybernetics
evolutionary theory
systems thinking
Where Clausewitz asks for vision, Boyd asks for learning.
c. Falsifiability and improvement
An OODA loop can fail:
bad observation
distorted orientation
slow decision
ineffective action
When it fails, the failure is diagnosable.
That is the decisive difference.
A unit, organization, or individual can:
shorten loops
improve sensing
diversify mental models
redesign feedback channels
OODA does not require genius. It assumes fallibility and designs around it.
That is the scientific posture.
===Dialectics vs. dynamics===
<pre>
Clausewitz: dialectical truth
Clausewitz’s worldview is Hegelian:
truth emerges from contradiction
war resists equilibrium
understanding is synthetic, not analytic
Coup d’œil is the mind’s ability to stand above contradiction without resolving it.
This is philosophy, not science.
Boyd: dynamical systems
Boyd’s worldview is Darwinian and cybernetic:
systems adapt or collapse
faster learning beats brute force
coherence is temporary
Victory comes not from transcendent insight but from continuous model destruction and renewal.
This is applied science.
5. Why this distinction matters
The difference is not academic—it predicts how each scales.
Dimension
Coup d’œil
OODA Loop
Authority
Individual
System
Replicability
No
Yes
Teachability
Limited
High
Debuggability
None
Strong
Suitability for AI
Poor
Excellent

Modern militaries, businesses, and AI systems overwhelmingly adopt Boyd, not Clausewitz, because:
machines cannot have mystical insight
but they can run loops
update models
and learn from error
6. Final claim (clean and sharp)
Coup d’œil is mystical because it locates truth inside the commander; OODA is scientific because it locates truth in the interaction between model and reality.
Clausewitz trusts vision.
Boyd trusts adaptation.
That is the fault line between pre‑modern strategic thought and modern systems thinking.
</pre>

==AI overtakes the web==

==References==
* See Wiki [[OODA]] for John Boyd's take on [[Dialectic Harmony]]

[[Category: Philosophy]]

LLM Training

2026-03-04T05:36:21Z

Tom: Created page with "==Meme== Training LLMs is a complex process divided into several key stages. Understanding these steps is crucial for building LLMs or improving upon existing models. ==Conte..."

==Meme==
Training LLMs is a complex process divided into several key stages. Understanding these steps is crucial for building LLMs or improving upon existing models.

==Context==
LLM training process

a. Data collection and preprocessing
The initial step in training an LLM is data collection and preparation. This data is sourced from various repositories, including scientific articles, websites, books, and curated databases. However, you need to clean and preprocess the raw information before using it to train the models. This involves removing unwanted noise, rectifying format inconsistencies, and discarding irrelevant details to prepare the data for tokenization.

During tokenization, the text is segmented into smaller, manageable units such as words or subwords. Techniques like Byte-Pair Encoding or WordPiece break down text into smaller pieces, called tokens, that the neural network uses for learning. These tokens form the foundation of the model's vocabulary and are critical for the model to start predicting language patterns.

b. Model configuration
After preprocessing the data, the next step is to configure the model, typically a transformer-based neural network due to its proven efficacy in natural language processing (NLP) tasks. Setting up this architecture involves specifying various parameters, such as the number of layers in the transformer blocks, the number of attention heads, and a variety of hyperparameters. Researchers extensively test different settings to find the best configuration that helps the model learn efficiently.

c. Model training
Once configuration is complete, the model enters the training phase where it’s exposed to the prepared text data. The model's objective is to predict the next word in a sequence, given the preceding words. When it predicts correctly, the model's internal weights are reinforced; when it errs, the weights are adjusted. This process contains countless iterations—spanning millions or even billions of instances across a huge dataset to refine the model's predictive capabilities.

d. Fine-tuning
As the model begins to form a basic understanding of language patterns from the extensive data it's exposed to, you can apply a higher level of structure to the model’s learning process with supervised fine-tuning (SFT). Instead of simply predicting the next word, the model learns to understand and generate responses aligned with human-provided examples. The SFT phase trains the model to provide accurate and contextually appropriate responses, which is essential for tools such as chatbots or virtual assistants. The model can undergo additional training with more data to improve its predictive accuracy and generalization capabilities.

Reinforcement learning from human feedback (RLHF) is an advanced stage of fine-tuning in which the model's responses are fine-tuned by comparing them with human preferences to ensure they are helpful, honest, and harmless (HHH). The model learns to prioritize better responses by creating various possible answers for a given input and having human labelers rank these from best to worst.

The RLHF process trains a reward model that understands the nuances of effective and ethical human conversation. Once trained, the reward model can replace human judgment at scale by providing feedback that helps tune the LLM to produce high-quality, valuable responses while avoiding misinformative or potentially harmful content.

In the next section, we'll delve deeper into the specific challenges presented by these stages and how they contribute to the broader complexities of LLM training.

Challenges of training LLMs

a. Data generation and validation
The data used for training LLMs must be carefully collected from diverse sources and then cleaned and organized into training, validation, and testing sets. Additionally, it needs to be reviewed for representation, bias, and ethical soundness. Building a data pipeline with the right labels and annotations is a time- and resource-intensive process, which makes it challenging for researchers and practitioners.

Here are some of the ways to approach data generation and validation concerns:

Creative strategies for dataset expansion: Employ techniques like back-translation, word replacement, and sentence shuffling to add richness and variety to your existing data.
Utilize available resources: Public datasets relevant to your task can serve as a solid starting point. They are freely accessible and can reduce the burden of initial data collection.
Crowdsourcing and expert review: Leverage online platforms to engage with different people who can help collect and label data to increase the diversity of your training dataset. Additionally, using human reviewers to monitor and moderate model’s outputs can ensure the model learns from human feedback and generates the desired output.
Enhancing model integrity: Implement strict data filtering and preprocessing routines to eliminate any offensive or inappropriate content. Incorporating adversarial training methods that expose the model to challenging examples can enhance its resilience against toxic inputs. Additionally, you can avoid biases by fine-tuning the model using curated datasets.
b. Optimizing reasoning capabilities
Developing LLMs that can process information and reason with humanlike logic remains one of the most advanced challenges in AI today. Models can understand and produce logic by recognizing language patterns, but true reasoning involves combining facts to draw conclusions.

One focused approach is to teach models the intricacies of coding. Coding represents a structured way of thinking, where logical operations are understood and applied in a precise sequence. This teaches the model systematic thinking, helping it to identify errors and figure out how to fix them, much like a coder debugging a program.

Another approach is RLHF which can boost a model's reasoning by incorporating human feedback into its learning process. "Chain-of-thought" prompting is another method that guides LLMs in the reasoning process by following logic examples.

DeepMind, for example, built a "faithful reasoning" framework for question-answering systems. It uses two specialized LLMs, one for selecting accurate information and another for inference, to construct a solid chain of logic. Each step in the process leads to the creation of a valid reasoning trace. For tasks that require multi-step deduction and scientific question answering, DeepMind's framework exceeds basic models in providing accurate final answers and generates easy-to-understand reasoning traces.

Despite these efforts, LLMs may falter as complexity grows, such as with natural language arithmetic questions. Large-scale neuro-symbolic AI systems like AI21 Labs' MRKL use a modular approach to reasoning that combines knowledge, computation, and natural language expression. This approach takes advantage of the full power of modern LLMs while overcoming their limitations, such as outdated or exclusive data and shortcomings in symbolic reasoning.

Finding effective ways to optimize reasoning capabilities, particularly through coding, can be extrapolated to broader logical applications. This enhances the model's problem-solving abilities and establishes a level of trust and reliability in the LLMs' capabilities.

c. Bias and hallucinations
Training LLMs presents two key challenges that can affect the model’s reliability and utility: bias and hallucinations. Even with extensive training on vast datasets, LLMs can still mimic deeply rooted biases present in the data and generate outputs that may reflect societal biases and distort the model's responses. Biases in AI can stem from various sources, including non-representative data reflecting societal norms, selection bias from limited data varieties, algorithmic bias from flawed processing, and biases rooted in race, culture, and politics. These biases can significantly alter how the AI understands and interprets the world.

Bias in LLMs can be eliminated by:

conducting thorough audits of the training datasets for equitable representation.
developing and applying algorithms that neutralize latent biases in data.
regular monitoring and iterative updates to the model to rectify any newly identified biases.
Hallucinations happen when LLMs get things wrong—producing information that diverges from reality. This can happen if the AI gets confused by the context or if there's missing information. These false outputs are incorrect and can be misleading when the stakes are high.

Hallucinations in LLMs can be prevented by:

improving the neural network’s architecture for better context processing.
enhancing the training data to provide a more accurate basis for predictions.
involving human judgment in evaluating LLM output, especially for sensitive use cases.
By addressing bias and hallucinations, companies can create models that are efficient, fair, and reliable—qualities essential for the responsible long-term use of the technology.

Want ethical AI that works in the real world?
Ensure Model Safety with Turing
d. Quality control and monitoring
Building a robust quality control setup is tough, whether you're tweaking an existing model or building one from scratch. The success of an LLM largely depends on its training data.; the model can only fulfill its full potential if the data is carefully preprocessed, diverse, and representative. This approach has proven successful by BloombergGPT, which was trained using a curated dataset consisting of 51 percent financial data and 49 percent general data. This strategy led to the model’s remarkable performance in financial tasks without compromising general LLM benchmarks.

Similarly, the Med-PaLM 2 model showcases the importance of quality input and stringent review by becoming the first LLM to pass the U.S. Medical Licensing Examination (USMLE), with a score of 67.2 percent on the MedQA dataset. The model underwent rigorous human evaluations that particularly focused on its clinical utility by assessing long-form medical questions across various dimensions.

Here’s a three-step approach to setting up effective quality control and monitoring:

Establish the right evaluation metrics: These metrics serve as benchmarks during and after training to ensure the LLM meets the defined quality standards. Some of the common metrics used to evaluate LLMs include:

a. Perplexity quantifies how uncertain a model is about the predictions it makes or measures the model's confidence in generating the next part of the text. A lower perplexity score indicates that the model is better at predicting the sample.

b. ROUGE (recall-oriented understudy for gisting evaluation) compares the overlap of word sequences between the output of the LLM and a human-produced reference text. This metric gauges the quality of machine-generated summaries or translations in terms of their alignment with human expectations.

c. F1 Score measures a model’s accuracy by considering both precision (how many selected items are relevant) and recall (how many relevant items are selected). This score is useful when it's crucial to manage both false positives and false negatives effectively.

How does your model measure up?
Evaluate Your Model Now
Continuous evaluation and iteration: Throughout training, the model is continuously assessed and fine-tuned to ensure each iteration brings the model closer to the desired level of performance. You can use libraries like Evaluate and TensorFlow Model Analysis (TFMA) for evaluating models and datasets.

Post-training review and feedback integration: Post-training evaluations, complemented by feedback from users, provide insights for refining the LLMs to make it robust and relevant.

Despite the importance of manual review in maintaining quality, there's a push to strike the right balance between using automated systems and human oversight. This balance is crucial for achieving scalable, efficient, and high-quality LLMs that can be integrated into real-world applications.

e. Technical expertise and operations management
Training and deploying LLMs requires a deep understanding of the technical aspects, such as deep-learning algorithms and transformers. Furthermore, it requires proficiency in coordinating the activities of numerous GPUs and managing the vast software and hardware intricacies.

Scaling teams rapidly: The ongoing evolution of LLM training projects and new workstreams demands the rapid scaling of teams with skilled LLM trainers and domain experts to maintain an edge in GenAI development. Training LLMs requires professionals with expertise in complex algorithms, model architectures, data cleaning, and model fine-tuning. However, due to the recent advancements in LLM technology, there's still an evident gap between the available skilled workforce and the growing demand that results in competitive hiring environments.

LLM training services have grown increasingly valuable by filling the gaps with the expertise needed to train, deploy, and maintain advanced LLMs. Beyond hiring, it's important to train existing employees on AI and ensure they know how to effectively use the tools, understand the limitations of LLMs, and integrate them into everyday tasks. Business leaders play a crucial role in this process by creating clear usage policies and offer ongoing training to maintain proficiency with these technologies.
Managing the operations: Running an LLM training and development team means investing in the support and guidance of each team member. It involves training developers, resolving their queries, overseeing output quality, and keeping them invested in their work. A developer success team is key to keeping teams motivated and productive by evaluating performance and encouraging a culture that promotes independent exploration and adoption of new AI tools in business processes.
Wrapping up
LLM training is a complex process that demands precision, innovative strategies, and a proactive approach to technology's ever-evolving landscape. Reflecting on the current limitations of LLMs, Yann LeCun, a Turing Award-winning machine learning researcher, observes, “It is clear that these models are doomed to a shallow understanding that will never approximate the full-bodied thinking we see in humans.” It’s critical to recognize the limitations of what AI can and cannot know through language alone. Companies must overcome biases, strengthen data quality, enhance reasoning skills, and manage both technical operations and talent to leverage the full potential of LLMs.
==References==

[[Category: Artificial Intelligence]]

Abuse Case Model

2026-03-04T05:17:59Z

Tom: /* Example */

==Definition==
An [[Abuse Case Model]] is built as an extension of a [[Use Case]].
==Example==
Identifying abuse before designing architecture: Embedding Game-Based Threat Modelling into Agile Delivery at a Major Latin American Bank

CyberSec Games March 3, 2026

This case study is based on the account of Max Alejandro Gómez-Sánchez Vergaray, AppSec & DevSecOps Consultant, who designed and led an application security programme during the digital transformation of one of the largest banks in Peru. Over the course of the programme, more than 3,000 people were trained in secure software development using [https://cornucopia.owasp.org/about OWASP Cornucopia].

Why Security Became Urgent
When I began leading the application security program, teams were delivering software at scale and the bank was undergoing digital transformation. Agile methods were well established but while security existed on paper it was not yet embedded into everyday engineering practice and lacked sponsorship.

Then the bank experienced a cyber attack.

Delivery teams paused feature work to resolve security problems and executive attention followed. External consultants were engaged at significant cost, and a threat modelling approach was introduced based on Trike - a semi-automated formal model focused on Denial of Service and Elevation of Privilege threats.

While the methodology was sound, the problem was sustainability. Consultancy-led threat modelling can assess systems, but it does not automatically create internal engineering capability. My goal was to ensure that security thinking became part of how teams built software, not something triggered only by incident or audit.

The Structural Problem in Agile
In a traditional SDLC, security activities are distributed across phases. Requirements analysis, design, development, and testing are each linked to security activities and checkpoints.

Agile changes that dynamic. Work begins with high-level user story exploration. Stories are refined and prioritised in sprint planning. Architecture often evolves in parallel. By the time a traditional threat model is created around an architectural diagram, many functional decisions have already been made.

In practice, identifying security requirements during architectural threat modelling is often too late. The backlog is already shaped. Trade-offs have been made. Time pressure is real.

I realized that if we wanted security to influence design meaningfully, it had to be integrated before architecture crystallized.

The Two Concepts That Changed the Model
To make this work, I relied on two foundational concepts: abuse case modelling and Secure Scrum.

Abuse case modelling takes each legitimate use case and models a corresponding negative scenario, behavior undesired by the system owner. The goal is not simply to enumerate threats, but to identify the security requirements that prevent those undesirable outcomes.

For example, if a use case involves user authentication, an abuse case might explore credential stuffing or brute force attempts. The discussion then naturally leads to security requirements such as rate limiting, password strength policies, or multi-factor enforcement.

Secure Scrum builds on this by integrating security requirements directly into user stories. Instead of treating security as an external checklist, security requirements become part of system requirements from the moment user stories are identified.

Why I Chose OWASP Cornucopia
I had experience with OWASP’s research and its card-based framework, OWASP Cornucopia. Traditionally, Cornucopia is played against an architectural diagram to identify threats.

I decided to reposition it.

The bank operated using a SAFe-aligned model with quarterly planning events so that an overall vision, work items, and dependencies are all identified before sprints begin for the quarter. These sessions brought cross-functional teams together to define plans and break them down into epics and stories

This was the ideal point to insert OWASP Cornucopia.

We shuffled the cards and to participants, ensuring a mix of suits. A facilitator guided the session, clarifying complex cards and moderating discussions. Epics were placed visibly on the wall. Each participant examined their cards and argued how specific threats or misuse scenarios related to particular epics or user stories.

The real power was not found in the mechanics of the game, but in the conversations it triggered. Engineers and Product Owners collectively explored what could go wrong and reasoned about attacker behavior in the context of real features.

From those discussions, we derived security requirements and attached them directly to backlog items.

Moving the “Magic” Earlier
In many Agile environments, threat modelling occurs after architecture diagrams are drafted. By then, the cost of change is higher.

By integrating Cornucopia into quarterly planning, we moved the magic earlier. We “spread left” as far as we could, to the very first artifacts in our agile workflow.

As a result, security requirements were identified well before sprint work began. When architecture was later discussed, it already reflected security-conscious decisions. Subsequent structured threat modelling sessions were more focused and more effective because they operated on stronger inputs.

We also aligned our requirements with recognized standards such as the OWASP Application Security Verification Standard. This ensured that our approach was not informal or purely cultural. It was grounded in established verification practices.

To support governance, we introduced a lightweight traceability matrix linking each user story to its associated security requirements and related weaknesses. This allowed us to manage defects systematically if requirements were not properly implemented, while keeping overhead manageable.

Scaling Capability
What began with a single team expanded steadily. Engineers appreciated that the method fit their workflow rather than disrupting it. Over time, I trained more than 3,000 people across the organization in secure software development practices using Cornucopia as an entry point.

Eventually, I implemented a train-the-trainer model. Security could not depend on one facilitator. Engineering leads were trained to run their own sessions. This distributed ownership was critical to sustainability.

Annual retraining ensured that security culture did not fade. Teams developed shared vocabulary around misuse, abuse cases, and preventive controls. Security thinking became habitual rather than exceptional.

The Real Value
The greatest value was not the number of threats identified. It was the change in behavior.

Security requirements became visible within user stories. Product Owners understood the impact of prioritization decisions. Architecture discussions included security trade-offs as standard practice. Late-stage rework decreased because risks surfaced earlier.

In recent years, automated and AI-driven threat modelling tools have become more common. Automation is valuable, but tools do not replace shared reasoning. When engineers collectively analyze how their systems could be abused, they develop judgement. That judgement influences countless micro-decisions that no tool can fully predict.

I often share a piece of advice I once received: it is much better to have a good threat model that’s finished than a perfect threat model that isn’t finished.

In Agile environments, pragmatism matters.
==Vulnerabilities==
==Solution==
# Start with a use case.

==References==
* See the wiki [[Threat Model]] for more details on that subject.

[[Category: Vulnerability]]

Abuse Case Model

2026-03-04T01:54:44Z

Tom: /* Example */

==Definition==
An [[Abuse Case Model]] is built as an extension of a [[Use Case]].
==Example==
Identifying abuse before designing architecture: Embedding Game-Based Threat Modelling into Agile Delivery at a Major Latin American Bank

CyberSec Games March 3, 2026

This case study is based on the account of Max Alejandro Gómez-Sánchez Vergaray, AppSec & DevSecOps Consultant, who designed and led an application security programme during the digital transformation of one of the largest banks in Peru. Over the course of the programme, more than 3,000 people were trained in secure software development using [https://cornucopia.owasp.org/about OWASP Cornucopia].

Why Security Became Urgent
When I began leading the application security program, teams were delivering software at scale and the bank was undergoing digital transformation. Agile methods were well established but while security existed on paper it was not yet embedded into everyday engineering practice and lacked sponsorship.

Then the bank experienced a cyber attack.

Delivery teams paused feature work to resolve security problems and executive attention followed. External consultants were engaged at significant cost, and a threat modelling approach was introduced based on Trike - a semi-automated formal model focused on Denial of Service and Elevation of Privilege threats.

While the methodology was sound, the problem was sustainability. Consultancy-led threat modelling can assess systems, but it does not automatically create internal engineering capability. My goal was to ensure that security thinking became part of how teams built software, not something triggered only by incident or audit.

The Structural Problem in Agile
In a traditional SDLC, security activities are distributed across phases. Requirements analysis, design, development, and testing are each linked to security activities and checkpoints.

Agile changes that dynamic. Work begins with high-level user story exploration. Stories are refined and prioritised in sprint planning. Architecture often evolves in parallel. By the time a traditional threat model is created around an architectural diagram, many functional decisions have already been made.

In practice, identifying security requirements during architectural threat modelling is often too late. The backlog is already shaped. Trade-offs have been made. Time pressure is real.

I realized that if we wanted security to influence design meaningfully, it had to be integrated before architecture crystallized.

The Two Concepts That Changed the Model
To make this work, I relied on two foundational concepts: abuse case modelling and Secure Scrum.

Abuse case modelling takes each legitimate use case and models a corresponding negative scenario, behavior undesired by the system owner. The goal is not simply to enumerate threats, but to identify the security requirements that prevent those undesirable outcomes.

For example, if a use case involves user authentication, an abuse case might explore credential stuffing or brute force attempts. The discussion then naturally leads to security requirements such as rate limiting, password strength policies, or multi-factor enforcement.

Secure Scrum builds on this by integrating security requirements directly into user stories. Instead of treating security as an external checklist, security requirements become part of system requirements from the moment user stories are identified.

Why I Chose OWASP Cornucopia
I had experience with OWASP’s research and its card-based framework, OWASP Cornucopia. Traditionally, Cornucopia is played against an architectural diagram to identify threats.

I decided to reposition it.

The bank operated using a SAFe-aligned model with quarterly planning events so that an overall vision, work items, and dependencies are all identified before sprints begin for the quarter. These sessions brought cross-functional teams together to define plans and break them down into epics and stories

This was the ideal point to insert OWASP Cornucopia.

We shuffled the cards and to participants, ensuring a mix of suits. A facilitator guided the session, clarifying complex cards and moderating discussions. Epics were placed visibly on the wall. Each participant examined their cards and argued how specific threats or misuse scenarios related to particular epics or user stories.

The real power was not found in the mechanics of the game, but in the conversations it triggered. Engineers and Product Owners collectively explored what could go wrong and reasoned about attacker behavior in the context of real features.

From those discussions, we derived security requirements and attached them directly to backlog items.

Moving the “Magic” Earlier
In many Agile environments, threat modelling occurs after architecture diagrams are drafted. By then, the cost of change is higher.

By integrating Cornucopia into quarterly planning, we moved the magic earlier. We “spread left” as far as we could, to the very first artifacts in our agile workflow.

As a result, security requirements were identified well before sprint work began. When architecture was later discussed, it already reflected security-conscious decisions. Subsequent structured threat modelling sessions were more focused and more effective because they operated on stronger inputs.

We also aligned our requirements with recognized standards such as the OWASP Application Security Verification Standard. This ensured that our approach was not informal or purely cultural. It was grounded in established verification practices.

To support governance, we introduced a lightweight traceability matrix linking each user story to its associated security requirements and related weaknesses. This allowed us to manage defects systematically if requirements were not properly implemented, while keeping overhead manageable.

Scaling Capability
What began with a single team expanded steadily. Engineers appreciated that the method fit their workflow rather than disrupting it. Over time, I trained more than 3,000 people across the organization in secure software development practices using Cornucopia as an entry point.

Eventually, I implemented a train-the-trainer model. Security could not depend on one facilitator. Engineering leads were trained to run their own sessions. This distributed ownership was critical to sustainability.

Annual retraining ensured that security culture did not fade. Teams developed shared vocabulary around misuse, abuse cases, and preventive controls. Security thinking became habitual rather than exceptional.

The Real Value
The greatest value was not the number of threats identified. It was the change in behavior.

Security requirements became visible within user stories. Product Owners understood the impact of prioritization decisions. Architecture discussions included security trade-offs as standard practice. Late-stage rework decreased because risks surfaced earlier.

In recent years, automated and AI-driven threat modelling tools have become more common. Automation is valuable, but tools do not replace shared reasoning. When engineers collectively analyze how their systems could be abused, they develop judgement. That judgement influences countless micro-decisions that no tool can fully predict.

I often share a piece of advice I once received: it is much better to have a good threat model that’s finished than a perfect threat model that isn’t finished.

In Agile environments, pragmatism matters. By e

==References==
* See the wiki [[Threat Model]] for more details on that subject.

[[Category: Vulnerability]]

Abuse Case Model

2026-03-04T00:11:01Z

Tom: /* Definition */

==Definition==
An [[Abuse Case Model]] is built as an extension of a [[Use Case]].
==Example==
Identifying abuse before designing architecture: Embedding Game-Based Threat Modelling into Agile Delivery at a Major Latin American Bank

CyberSec Games March 3, 2026

This case study is based on the account of Max Alejandro Gómez-Sánchez Vergaray, AppSec & DevSecOps Consultant, who designed and led an application security programme during the digital transformation of one of the largest banks in Peru. Over the course of the programme, more than 3,000 people were trained in secure software development using OWASP Cornucopia.

Why Security Became Urgent
When I began leading the application security programme, teams were delivering software at scale and the bank was undergoing digital transformation. Agile methods were well established but while security existed on paper it was not yet embedded into everyday engineering practice and lacked sponsorship.

Then the bank experienced a cyber attack.

Delivery teams paused feature work to resolve security problems and executive attention followed. External consultants were engaged at significant cost, and a threat modelling approach was introduced based on Trike - a semi-automated formal model focused on Denial of Service and Elevation of Privilege threats.

While the methodology was sound, the problem was sustainability. Consultancy-led threat modelling can assess systems, but it does not automatically create internal engineering capability. My goal was to ensure that security thinking became part of how teams built software, not something triggered only by incident or audit.

The Structural Problem in Agile
In a traditional SDLC, security activities are distributed across phases. Requirements analysis, design, development, and testing are each linked to security activities and checkpoints.

Agile changes that dynamic. Work begins with high-level user story exploration. Stories are refined and prioritised in sprint planning. Architecture often evolves in parallel. By the time a traditional threat model is created around an architectural diagram, many functional decisions have already been made.

In practice, identifying security requirements during architectural threat modelling is often too late. The backlog is already shaped. Trade-offs have been made. Time pressure is real.

I realised that if we wanted security to influence design meaningfully, it had to be integrated before architecture crystallised.

The Two Concepts That Changed the Model
To make this work, I relied on two foundational concepts: abuse case modelling and Secure Scrum.

Abuse case modelling takes each legitimate use case and models a corresponding negative scenario, behaviour undesired by the system owner. The goal is not simply to enumerate threats, but to identify the security requirements that prevent those undesirable outcomes.

For example, if a use case involves user authentication, an abuse case might explore credential stuffing or brute force attempts. The discussion then naturally leads to security requirements such as rate limiting, password strength policies, or multi-factor enforcement.

Secure Scrum builds on this by integrating security requirements directly into user stories. Instead of treating security as an external checklist, security requirements become part of system requirements from the moment user stories are identified.

Why I Chose OWASP Cornucopia
I had experience with OWASP’s research and its card-based framework, OWASP Cornucopia. Traditionally, Cornucopia is played against an architectural diagram to identify threats.

I decided to reposition it.

The bank operated using a SAFe-aligned model with quarterly planning events so that an overall vision, work items, and dependencies are all identified before sprints begin for the quarter. These sessions brought cross-functional teams together to define plans and break them down into epics and stories

This was the ideal point to insert OWASP Cornucopia.

We shuffled the cards and to participants, ensuring a mix of suits. A facilitator guided the session, clarifying complex cards and moderating discussions. Epics were placed visibly on the wall. Each participant examined their cards and argued how specific threats or misuse scenarios related to particular epics or user stories.

The real power was not found in the mechanics of the game, but in the conversations it triggered. Engineers and Product Owners collectively explored what could go wrong and reasoned about attacker behaviour in the context of real features.

From those discussions, we derived security requirements and attached them directly to backlog items.

Moving the “Magic” Earlier
In many Agile environments, threat modelling occurs after architecture diagrams are drafted. By then, the cost of change is higher.

By integrating Cornucopia into quarterly planning, we moved the magic earlier. We “spread left” as far as we could, to the very first artifacts in our agile workflow.

As a result, security requirements were identified well before sprint work began. When architecture was later discussed, it already reflected security-conscious decisions. Subsequent structured threat modelling sessions were more focused and more effective because they operated on stronger inputs.

We also aligned our requirements with recognised standards such as the OWASP Application Security Verification Standard. This ensured that our approach was not informal or purely cultural. It was grounded in established verification practices.

To support governance, we introduced a lightweight traceability matrix linking each user story to its associated security requirements and related weaknesses. This allowed us to manage defects systematically if requirements were not properly implemented, while keeping overhead manageable.

Scaling Capability
What began with a single team expanded steadily. Engineers appreciated that the method fit their workflow rather than disrupting it. Over time, I trained more than 3,000 people across the organisation in secure software development practices using Cornucopia as an entry point.

Eventually, I implemented a train-the-trainer model. Security could not depend on one facilitator. Engineering leads were trained to run their own sessions. This distributed ownership was critical to sustainability.

Annual retraining ensured that security culture did not fade. Teams developed shared vocabulary around misuse, abuse cases, and preventive controls. Security thinking became habitual rather than exceptional.

The Real Value
The greatest value was not the number of threats identified. It was the change in behaviour.

Security requirements became visible within user stories. Product Owners understood the impact of prioritisation decisions. Architecture discussions included security trade-offs as standard practice. Late-stage rework decreased because risks surfaced earlier.

In recent years, automated and AI-driven threat modelling tools have become more common. Automation is valuable, but tools do not replace shared reasoning. When engineers collectively analyse how their systems could be abused, they develop judgement. That judgement influences countless micro-decisions that no tool can fully predict.

I often share a piece of advice I once received: it is much better to have a good threat model that’s finished than a perfect threat model that isn’t finished.

In Agile environments, pragmatism matters. By e

Abuse Case Model

2026-03-04T00:04:38Z

Tom: Created page with "==Definition== Identifying abuse before designing architecture: Embedding Game-Based Threat Modelling into Agile Delivery at a Major Latin American Bank CyberSec Games 510 fo..."

==Definition==
Identifying abuse before designing architecture: Embedding Game-Based Threat Modelling into Agile Delivery at a Major Latin American Bank

CyberSec Games
510 followers

March 3, 2026
This case study is based on the account of Max Alejandro Gómez-Sánchez Vergaray, AppSec & DevSecOps Consultant, who designed and led an application security programme during the digital transformation of one of the largest banks in Peru. Over the course of the programme, more than 3,000 people were trained in secure software development using OWASP Cornucopia.

Why Security Became Urgent
When I began leading the application security programme, teams were delivering software at scale and the bank was undergoing digital transformation. Agile methods were well established but while security existed on paper it was not yet embedded into everyday engineering practice and lacked sponsorship.

Then the bank experienced a cyber attack.

Delivery teams paused feature work to resolve security problems and executive attention followed. External consultants were engaged at significant cost, and a threat modelling approach was introduced based on Trike - a semi-automated formal model focused on Denial of Service and Elevation of Privilege threats.

While the methodology was sound, the problem was sustainability. Consultancy-led threat modelling can assess systems, but it does not automatically create internal engineering capability. My goal was to ensure that security thinking became part of how teams built software, not something triggered only by incident or audit.

The Structural Problem in Agile
In a traditional SDLC, security activities are distributed across phases. Requirements analysis, design, development, and testing are each linked to security activities and checkpoints.

Agile changes that dynamic. Work begins with high-level user story exploration. Stories are refined and prioritised in sprint planning. Architecture often evolves in parallel. By the time a traditional threat model is created around an architectural diagram, many functional decisions have already been made.

In practice, identifying security requirements during architectural threat modelling is often too late. The backlog is already shaped. Trade-offs have been made. Time pressure is real.

I realised that if we wanted security to influence design meaningfully, it had to be integrated before architecture crystallised.

The Two Concepts That Changed the Model
To make this work, I relied on two foundational concepts: abuse case modelling and Secure Scrum.

Abuse case modelling takes each legitimate use case and models a corresponding negative scenario, behaviour undesired by the system owner. The goal is not simply to enumerate threats, but to identify the security requirements that prevent those undesirable outcomes.

For example, if a use case involves user authentication, an abuse case might explore credential stuffing or brute force attempts. The discussion then naturally leads to security requirements such as rate limiting, password strength policies, or multi-factor enforcement.

Secure Scrum builds on this by integrating security requirements directly into user stories. Instead of treating security as an external checklist, security requirements become part of system requirements from the moment user stories are identified.

Why I Chose OWASP Cornucopia
I had experience with OWASP’s research and its card-based framework, OWASP Cornucopia. Traditionally, Cornucopia is played against an architectural diagram to identify threats.

I decided to reposition it.

The bank operated using a SAFe-aligned model with quarterly planning events so that an overall vision, work items, and dependencies are all identified before sprints begin for the quarter. These sessions brought cross-functional teams together to define plans and break them down into epics and stories

This was the ideal point to insert OWASP Cornucopia.

We shuffled the cards and to participants, ensuring a mix of suits. A facilitator guided the session, clarifying complex cards and moderating discussions. Epics were placed visibly on the wall. Each participant examined their cards and argued how specific threats or misuse scenarios related to particular epics or user stories.

The real power was not found in the mechanics of the game, but in the conversations it triggered. Engineers and Product Owners collectively explored what could go wrong and reasoned about attacker behaviour in the context of real features.

From those discussions, we derived security requirements and attached them directly to backlog items.

Moving the “Magic” Earlier
In many Agile environments, threat modelling occurs after architecture diagrams are drafted. By then, the cost of change is higher.

By integrating Cornucopia into quarterly planning, we moved the magic earlier. We “spread left” as far as we could, to the very first artifacts in our agile workflow.

As a result, security requirements were identified well before sprint work began. When architecture was later discussed, it already reflected security-conscious decisions. Subsequent structured threat modelling sessions were more focused and more effective because they operated on stronger inputs.

We also aligned our requirements with recognised standards such as the OWASP Application Security Verification Standard. This ensured that our approach was not informal or purely cultural. It was grounded in established verification practices.

To support governance, we introduced a lightweight traceability matrix linking each user story to its associated security requirements and related weaknesses. This allowed us to manage defects systematically if requirements were not properly implemented, while keeping overhead manageable.

Scaling Capability
What began with a single team expanded steadily. Engineers appreciated that the method fit their workflow rather than disrupting it. Over time, I trained more than 3,000 people across the organisation in secure software development practices using Cornucopia as an entry point.

Eventually, I implemented a train-the-trainer model. Security could not depend on one facilitator. Engineering leads were trained to run their own sessions. This distributed ownership was critical to sustainability.

Annual retraining ensured that security culture did not fade. Teams developed shared vocabulary around misuse, abuse cases, and preventive controls. Security thinking became habitual rather than exceptional.

The Real Value
The greatest value was not the number of threats identified. It was the change in behaviour.

Security requirements became visible within user stories. Product Owners understood the impact of prioritisation decisions. Architecture discussions included security trade-offs as standard practice. Late-stage rework decreased because risks surfaced earlier.

In recent years, automated and AI-driven threat modelling tools have become more common. Automation is valuable, but tools do not replace shared reasoning. When engineers collectively analyse how their systems could be abused, they develop judgement. That judgement influences countless micro-decisions that no tool can fully predict.

I often share a piece of advice I once received: it is much better to have a good threat model that’s finished than a perfect threat model that isn’t finished.

In Agile environments, pragmatism matters. By e

Dark Matter

2026-03-03T00:11:13Z

Tom: /* But What About */

==Full Title==

Dark matter is a mysterious substance that makes up most of the matter in the universe, but does not interact with light or other electromagnetic radiation. Scientists have inferred its existence from its gravitational effects on visible matter, such as stars and galaxies.<ref>Britanica ''Dark Matter'' https://www.britannica.com/science/dark-matter</ref>
==What the Scientists Say==
There are different theories about what dark matter is made of, but none of them have been confirmed by direct observation. Some possible candidates are:

*Weakly interacting massive particles (WIMPs): These are hypothetical particles that only interact through gravity and the weak nuclear force. They are predicted by some extensions of the standard model of particle physics, such as supersymmetry
* Axions: These are hypothetical particles that are very light and very weakly interacting. They were proposed to solve a problem in quantum chromodynamics, the theory of the strong nuclear force<ref name=nasa>NASA ''What is Dark Matter'' https://www.nasa.gov/audience/forstudents/9-12/features/what-is-dark-matter.html</ref>
* Primordial black holes: These are black holes that formed in the early universe, before stars and galaxies. They could have a wide range of masses and sizes, depending on the conditions of their formation<ref name=nasa />

Scientists are trying to detect dark matter particles directly using various experiments, such as underground detectors, space telescopes, and particle colliders. However, so far none of these experiments have found conclusive evidence for dark matter<ref>ESA ''What is Dark Matter'' https://sci.esa.int/web/euclid/-/what-is-dark-matter- </ref>

Dark matter is one of the biggest mysteries in modern physics and cosmology. It has profound implications for our understanding of the origin, structure, and fate of the universe.

==But What About==
The formula in General Relativity is just wrong. After all, its only be tested on nearby objects (the size of the solar system) where we understand how time works.
==Fun==
This is just for sharing a previous translated chat, no action required. HOW YOUR FINGERS COULD GO THROUGH A TABLE
===Can the electrons in my finger exchange with the electrons in the table I’m pressing?===
In ordinary conditions: no. Even though electrons are identical quantum particles, the reason they do not “exchange” in any noticeable way is: * The Pauli exclusion principle prevents electrons from occupying the same quantum state. * The electromagnetic repulsion between atoms creates a large effective barrier. * What we feel as “solid contact” is actually electron cloud repulsion. At a deeper quantum level: * All electrons are excitations of the same underlying electron field. * But exchanging electrons between objects would require overlapping quantum states and tunneling probabilities that are astronomically small. So practically speaking: it does not happen.
===In what theoretical situation could the atoms of my finger pass through the table?===
This would require overcoming the electromagnetic repulsion barrier between atoms. The only known mechanism is **quantum tunneling**. However: * For a single electron, tunneling through a small barrier is possible. * For an entire atom, the probability becomes extremely tiny. * For a macroscopic object like a finger (≈10²⁷ atoms), the probability is effectively zero. The tunneling probability decreases exponentially with: * Barrier height * Barrier width * Mass of the particle For macroscopic objects, the probability is so small it is practically impossible.
===What could increase tunneling probability? High electrostatics?===
Increasing electric fields generally increases the barrier, not reduces it. To increase tunneling probability you would need: * Lower barrier height * Narrower barrier * Smaller mass High electrostatics would typically increase repulsion. To make macroscopic tunneling plausible would require: * Extreme quantum coherence * Or altering fundamental constants * Or new physics
===What if all my atoms vibrated at a different frequency than the table’s?===
In classical terms, frequency mismatch does not remove electromagnetic interaction. Atoms interact because of charge and quantum fields, not because of synchronized vibration. Even if vibrational frequencies differed: * Coulomb forces still apply. * Electron clouds still repel. So frequency mismatch does not allow interpenetration.
===What about matter waves (de Broglie frequencies)?===
All particles have a de Broglie wavelength: [ \lambda = \frac{h}{p} ] But for macroscopic objects: * Momentum is enormous. * Wavelength is absurdly small. * Wave-like behavior is unobservable. Different de Broglie frequencies do not eliminate interaction.
===What about string theory — different vibrational frequencies?===
In string theory: * Particles are vibrational modes of strings. * Different modes correspond to different masses and charges. However: Interaction is not determined solely by vibrational frequency. It is determined by: * Gauge charges * Allowed couplings in the theory Even if two particles vibrate differently, if they share gauge charges, they interact.
===Why doesn’t dark matter interact with normal matter?===
Because it does not carry Standard Model gauge charges. Normal matter interacts through: * Electromagnetism (U(1)) * Weak force (SU(2)) * Strong force (SU(3)) Dark matter, in most models: * Has no charge under these groups. * Therefore cannot couple to photons, gluons, or W/Z bosons. * It may only interact gravitationally. Interaction requires coupling in the Lagrangian — not proximity.
===Invent a dark-matter-like version of my finger that overlaps but doesn’t interact.===
To construct this theoretically: * Introduce a second gauge sector. * Introduce a second U(1) gauge field (dark electromagnetism). * Introduce dark electrons and dark protons charged under dark U(1). * Form dark atoms. * Form dark chemistry. * Form dark macroscopic structures. These objects: * Would overlap spatially with normal matter. * But would not exchange photons. * Would not experience electromagnetic repulsion. * Would pass through normal matter. Only gravity would couple them.
===What is the most important fundamental difference?===
The most fundamental difference is: > Gauge charge structure. Normal particles: * Charged under Standard Model gauge groups. Dark particles: * Charged under a different gauge group. * Neutral under Standard Model gauge groups. Because of this: * No electromagnetic interaction. * No atomic repulsion. * No overlap resistance.
=== What is a gauge charge structure? Gauge charge structure is: > The set of symmetry groups under which particles carry charge and interact. For example: * U(1) → electromagnetism * SU(2) → weak interaction * SU(3) → strong interaction Particles interact if they share gauge charges.
===What is brane tension?===
In string theory: A brane is an extended object. Its tension is: > Energy per unit volume (or area). It represents intrinsic stored energy due to its existence as an extended object. It does not require vibration. It is analogous to energy stored in a stretched string.
=== What are extra dimensions? In string theory: Spacetime may have 10 or 11 dimensions. We observe 3 spatial dimensions. The others are compactified (curled up). Two branes can: * Overlap in our visible 3D * Be separated in an extra dimension Thus appear coincident but remain independent.
===Does energy require vibration?===
No. Energy can be: * Rest mass energy (mc²) * Static field energy * Vacuum energy * Topological energy * Tension energy Vibration is only one form of energy.
===Can an electron exist at rest?===
Yes. Relativistic rest means: * Momentum average = 0 Quantum mechanically: * Fluctuations always exist. * Exact stillness is impossible. * But the particle absolutely exists. It has rest energy: [ E_0 = mc^2 ]
===If it has quantum fluctuations, does it have a natural frequency?===
Yes. For stationary states: [ \psi(t) \sim e^{-iEt/\hbar} ] This implies a natural quantum frequency: [ \omega = \frac{E}{\hbar} ] So even with zero average motion, the quantum state evolves in phase. This is not classical vibration — it is phase evolution.
=== Would a dark electron vibrate differently?===
In string theory terms: A normal electron: * Open string attached to Standard Model brane. * Specific vibrational mode. A dark electron: * Open string attached to dark brane. * Different vibrational mode. * Different rest mass. * Different natural quantum frequency. * Charged under dark U(1), not our U(1). Thus: [ \omega_{\text{dark}} = \frac{E_{\text{dark}}}{\hbar} ] Different energy → different intrinsic frequency. But the deeper difference is gauge structure.

The reason matter does not overlap is: * Shared gauge structure * Electromagnetic repulsion * Pauli exclusion * Quantum interactions The only way to allow overlap without interaction is: > Different gauge sector + no coupling terms. Not frequency mismatch. Not classical vibration differences. Not de Broglie wavelength differences. The true separator is symmetry structure and charge.

If you want, we can now go one level deeper: Is reality fundamentally vibration? Or fundamentally quantum fields? Or fundamentally symmetry?

==References==

[[Category: Physics]]

Someone Else's Problem

2026-02-27T17:59:55Z

Tom: /* Solutions */

==Meme==
An **SEP field** makes an object effectively invisible by exploiting a mental blind spot:

* If something is perceived as **“somebody else’s problem”**, people simply **don’t notice it**.
* The object is still there, fully visible in a literal sense.
* Observers unconsciously assume it’s not their responsibility to understand, question, or deal with it—so their minds slide right past it.
==History==
In the Hitchhiker's Guide’s logic, and SEP field is *far more effective* than traditional invisibility.

===The *source* of the SEP field===

The SEP field is generated by a device commonly referred to as an **SEP field generator** (famously used on the starship **Heart of Gold**).

Conceptually, its “source” is:

* **Human (and alien) psychology**, not physics
* Specifically, the tendency to ignore:
* Bureaucracy
* Technical infrastructure
* Construction projects
* Things that look official, complicated, or already “handled”

Adams jokes that the SEP field is one of the cheapest and most powerful forms of invisibility ever invented, because it doesn’t require bending light or space = just **expectations**.

> If something looks like it belongs to someone else, the brain declines to engage.

=== What problem does it solve?===

The SEP field solves the problem of **being noticed at all**.

* A huge starship can sit in the middle of a cricket field.
* As long as it is plausibly “somebody else’s problem,” people will:
* Walk around it
* Not question it
* Not remember it later

This mirrors real life:

* Roadworks you never asked for
* Government forms you don’t understand
* Massive systems everyone assumes “someone else” is in charge of

===Why the idea matters (the joke behind it)===

Douglas Adams wasn’t really writing about force fields.

The SEP field is satire about:

* Bureaucracy
* Technology
* Institutional neglect
* Human avoidance behavior

It explains why:

* Enormous problems persist
* Everyone assumes someone else is handling them
* Nothing gets fixed until it affects you personally

===One‑sentence summary===
The SEP field works by convincing your brain that something isn’t your problem. It stops working the instant you decide that it is your problem.

==Solutions==
An SEP field **fails instantly** when one condition is met:

>> When it becomes *your* problem**

The moment someone:

* Takes responsibility
* Feels curiosity
* Is directly affected
* Or emotionally engages

…the field collapses for them.

In the books:

* **Arthur Dent** can sometimes see through things others can’t because he *doesn’t share the same assumptions* as the galactic norm.
* The Guide repeatedly implies that awareness, responsibility, or plain stubbornness defeats the effect.

In other words:

> **The solution to an SEP field is attention.**

Below is a clear comparison between the SEP field and real‑world psychology, mapped concept by concept.

===SEP Field ≈ Bystander Effect & Diffusion of Responsibility===

**In the books**

* The SEP field works because everyone assumes someone else is dealing with it.
* Responsibility feels *distributed*, so no one acts.

**In psychology**

* The **bystander effect** shows that people are less likely to intervene when others are present.
* **Diffusion of responsibility** means each person feels less personal obligation as group size increases.

Psychological research consistently shows that when responsibility is ambiguous or shared, inaction becomes the default.

**Why this maps perfectly**

* An SEP field doesn’t make action impossible.
* It makes action feel *unnecessary*—which is exactly how diffusion of responsibility operates.

> SEP field logic: “Someone else must be handling this.”
> Psychology: “Someone else will probably do it.”

===SEP Field ≈ Inattentional Blindness===

**In the books**

* People don’t just refuse to act—they often **don’t notice** the object at all.
* A massive spaceship can sit in plain sight.

**In psychology**

* **Inattentional blindness** is the failure to notice visible, salient objects because attention is focused elsewhere.
* Famous experiments show people missing a gorilla walking through a scene they are watching carefully.

This is not a vision problem; it’s an **attention allocation problem**.

**Why this maps perfectly**

* SEP fields exploit attention, not perception.
* If the brain decides something is irrelevant, it effectively vanishes from awareness.

> SEP field logic: “This doesn’t concern me.”
> Psychology: “This doesn’t enter my attentional set.”

===SEP Field ≈ Status Quo Bias===

**In the books**

* The SEP field keeps things the way they are.
* No one challenges the object’s presence or purpose.

**In psychology**

* **Status quo bias** is the preference for leaving things unchanged, even when change would be beneficial.
* Defaults are disproportionately powerful; people stick with them to avoid cognitive effort and potential loss.

Status quo bias is driven by loss aversion, cognitive load, and fear of uncertainty.

**Why this maps perfectly**

* Investigating an SEP‑covered object requires effort and risk.
* Ignoring it preserves comfort and mental efficiency.

> SEP field logic: “This must be normal.”
> Psychology: “Changing this would cost effort and risk.”

===SEP Field ≈ Bureaucratic and Institutional Neglect===

**In the books**

* SEP fields are especially effective on:
* Infrastructure
* Official projects
* Things that look “authorized”

**In real life**

* Large systems often fail through **bureaucratic neglect**, where:
* Responsibility is fragmented
* Accountability is unclear
* Problems persist because no single actor owns them

Research on administrative burden and bureaucratic systems shows how people explain failures by attributing them to “the system” rather than individuals, which reduces pressure to act.

**Why this maps perfectly**

* SEP fields scale *upward*: the bigger and more official something looks, the more invisible it becomes.
* Institutions unintentionally generate real SEP fields.

> SEP field logic: “This is clearly official business.”
> Psychology: “The system, not me, is responsible.”

===How SEP Fields Fail — Psychology Agrees===

In *Hitchhiker’s*, SEP fields collapse when:

* Someone takes personal responsibility
* The object directly affects them
* Curiosity or emotional engagement overrides apathy

Psychology shows the same pattern:

* The bystander effect weakens when **responsibility is made explicit**
* Inattentional blindness disappears when attention is redirected
* Status quo bias breaks when defaults are removed

In short, **personal ownership destroys the effect**.

## Bottom line

Douglas Adams didn’t invent a fake force field.

He described, with comic precision, what happens when:

* Responsibility is diffuse
* Attention is misdirected
* Systems feel larger than individuals
* Change feels costly

**The SEP field is what psychology looks like from the outside.**
==Compared to 1984==
Below is a clean, concept‑by‑concept comparison.

## Core difference (one sentence)

**The SEP field hides reality by making people ignore it; *1984* controls reality by making people see only what they’re told.**
===Attention vs. Enforcement===

### SEP field (Douglas Adams)

* Operates through **neglect**
* No force, no authority, no punishment
* People *could* notice the problem—but don’t
* Power comes from **apathy and cognitive avoidance**

The SEP field exploits the mental shortcut:

> *“This doesn’t concern me.”*

### *1984* (George Orwell)

* Operates through **coercion**
* Surveillance, punishment, indoctrination
* People notice reality but must **deny it**
* Power comes from **fear and enforced belief**

The Party enforces the rule:

> *“This concerns you—and we control what it means.”*

===Responsibility: Absent vs. Absolute===

### SEP field

* Responsibility is **diffuse**
* Everyone assumes someone else is handling it
* No one feels personally accountable
* Inaction is psychologically comfortable

This mirrors bureaucratic shrugging and large‑system inertia.

### *1984*

* Responsibility is **total and unavoidable**
* You are accountable for:
* Your words
* Your thoughts
* Your facial expressions
* Inaction is not safe; obedience is mandatory

Where the SEP field dissolves responsibility, *1984* weaponizes it.

===Truth: Ignored vs. Rewritten===

### SEP field

* Truth still exists
* It is simply **not engaged with**
* Reality remains intact but unexamined

The problem isn’t falsehood—it’s **disinterest**.

### *1984*

* Truth is **actively destroyed**
* Facts are rewritten
* Contradictions must be believed simultaneously (doublethink)

The problem isn’t ignorance—it’s **compulsory belief**.
===Visibility: Psychological Invisibility vs. Total Surveillance===

### SEP field

* Things are “invisible” because no one looks
* Visibility collapses due to **attention failure**
* No one is watching because no one cares

### *1984*

* Nothing is invisible
* Everyone is watched all the time
* Visibility is a tool of control

In SEP worlds, problems vanish into neglect.
In *1984*, people vanish because they are noticed.
===Failure mode of society===

### SEP field describes:

* Late‑stage bureaucracy
* Large organizations
* Infrastructure decay
* “Everyone assumes it’s handled”

This is the failure mode of **complex systems without ownership**.

### *1984* describes:

* Totalitarian states
* Ideological absolutism
* Fear‑based conformity
* Reality as a political weapon

This is the failure mode of **centralized power without limits**.

===How each system breaks===

### SEP field collapses when:

* Someone takes personal responsibility
* Curiosity overrides apathy
* A problem becomes *your* problem

> Attention defeats the SEP field.

### *1984* collapses only if:

* Fear is removed
* Surveillance fails
* Power loses its monopoly on violence and truth

> Courage alone is not enough; the system must crack.

===Why they are often confused (but shouldn’t be)===

Both explain why:

* Obvious injustices persist
* People don’t intervene
* Absurd realities become normalized

But:

| SEP field | *1984* |
| ---------------- | ------------------------ |
| Neglect | Oppression |
| Apathy | Terror |
| Inattention | Indoctrination |
| “Not my problem” | “Only the Party decides” |

They are **opposite ends of the same spectrum**.

## Final synthesis

Douglas Adams warns:

> *“The greatest danger is not tyranny—it’s assuming someone else will deal with it.”*

George Orwell warns:

> *“The greatest danger is tyranny deciding what you are allowed to see and think.”*

**SEP fields thrive where people look away.
*1984* thrives where people are forced to look and then lie.**

==References==

[[Category: Psychology]]

Someone Else's Problem

2026-02-27T17:44:50Z

Tom: /* The *solution* to an SEP field (how it’s defeated) */

Someone Else's Problem

2026-02-27T17:40:54Z

Tom: /* The *source* of the SEP field */

Someone Else's Problem

2026-02-27T17:38:58Z

Tom: /* One‑sentence summary */

Someone Else's Problem

2026-02-27T17:37:50Z

Tom: Created page with "==Meme== An **SEP field** makes an object effectively invisible by exploiting a mental blind spot: * If something is perceived as **“somebody else’s problem”**, peopl..."

Digital Rights Management

2026-02-27T17:33:39Z

Tom: /* Device Control */

==Full Title or Meme==
The control of access to content on computing devices that are not owned by the content owner.

==Context==
For individual users.
* Information wants to be free!
* Unless, of course, it might embarrass me or interfere with my ability to earn a living.
For owners of valuable resources.
* The realization of large media companies, like music and videos, saw their control of the distribution channels erode.
* The governments and militaries of the world would rather that the enemy not see their plans or even their knowledge.
* Any large organization similarly wants to control access to their internal resources as a source of market control.

After the failure of the media companies from 2000 onward to control access to their content, DRM was known by different names. But the goal remains the same as ever, Enterprises want to control access to information, even when it is on devices that are not under their control.

==Solutions==
===Device Control===
* At first the media companies attempted to convince the lawmakers that it was [[Someone Else's Problem]] (SEP) to protect their revenue stream. This SEP field worked for a while, but eventually they had to back off.
* Enterprise-owed devices allowed the Enterprise to install Mobile Device Management (MDM) on the device and continue to control their content no matter where the user would travel.
* Bring your Own Device (ByoD) to work allows some level of control, but there is resistance to allowing an Enterprise MDM on a user-owned device.
* The best solution has proven to be an [[Enterprise]] supplied app that allowed access to [Enterprise] data.

===Policy===
See the wiki page on [[Policy-Based Access Control]] for more examples of policy languages. This works best, of course, on apps owned or controlled by the [[Enterprise]].
* [https://www.oasis-open.org/committees/tc_home.php?wg_abbrev=xacml OASIS eXtensible Access Control Markup Language (XACML) TC] with errata Version 2 (2008-07-29), version 3 (2017-07-12). <blockquote>The [https://www.oasis-open.org/committees/tc_home.php?wg_abbrev=xacml XACML Technical Committee] defines a core XML schema for representing authorization and entitlement policies.</blockquote>
* [https://www.w3.org/TR/odrl-model/ 2018-02-15 ODRL Information Model 2 W3C Recommendation] <blockquote>Policies are used to represent permitted and prohibited actions over a certain asset, as well as the obligations required to be meet by stakeholders. In addition, policies may be limited by constraints (e.g., temporal or spatial constraints) and duties (e.g. payments) may be imposed on permissions.</blockquote>
* 2002-09-19 [https://www.w3.org/TR/odrl/#:~:text=The%20Open%20Digital%20Rights%20Language,expressing%20rights%20information%20over%20content.&text=The%20ODRL%20has%20no%20license,of%20%22open%20source%22%20software. Open Digital Rights Language (ODRL) Version 1.1] <blockquote>The Open Digital Rights Language (ODRL) is a proposed language for the Digital Rights Management (DRM) community for the standardization of expressing rights information over content. The ODRL is intended to provide flexible and interoperable mechanisms to support transparent and innovative use of digital resources in publishing, distributing and consuming of electronic publications, digital images, audio and movies, learning objects, computer software and other creations in digital form. The ODRL has no license requirements and is available in the spirit of "open source" software. </blockquote>

==References==

[[Category: Glossary]]
[[Category: Authorization]]
[[Category: Policy]]
[[Category: Language]]

Graph Theory

2026-02-25T19:32:58Z

Tom: /* Taxonomy */

==Meme==
[[Graph Theory]]
==Context==
[[Trust Graph]]
==Taxonomy==
# By Structure:
## Undirected Graph: Edges have no direction; connections are bidirectional
## Directed Graph (Digraph): Edges have direction; connections flow one way
## Weighted Graph: Edges have assigned values or costs
## Unweighted Graph: All edges are equal; no values assigned
# By Connectivity:
## Connected Graph: Path exists between every pair of vertices
## Disconnected Graph: Some vertices have no connecting path
## Complete Graph: Every vertex connects to every other vertex
## Tree: Connected graph with no cycles
# By Cycles:
## Acyclic Graph: Contains no cycles or loops
## Cyclic Graph: Contains at least one cycle
## DAG (Directed Acyclic Graph): Directed graph with no cycles
# By Representation:
## Simple Graph: No self-loops or multiple edges between vertices
## Multigraph: Multiple edges allowed between vertices
## Pseudograph: Self-loops and multiple edges allowed
# By Application:
## Social Network Graph: Represents relationships between people
## Knowledge Graph: Represents entities and their relationships
## Flow Network: Models flow of resources through a network
## Dependency Graph: Shows dependencies between elements
## [[Trust Graph]]: Allows evaluation or attestation of Entities (Nodes)

Node (Vertex): A fundamental point or entity in a graph that represents an object or data point

Edge (Link/Arc): A connection between two vertices that represents a relationship or path

Degree: The number of edges connected to a vertex

In-degree: Number of incoming edges (directed graphs)
* Out-degree: Number of outgoing edges (directed graphs)
* Weight: A numerical value assigned to an edge representing cost, distance, or strength

Path: A sequence of vertices connected by edges

Tree: A connected, acyclic graph where any two vertices are connected by exactly one path

Cycle: A path that starts and ends at the same vertex

Loop (Self-loop): An edge that connects a vertex to itself

Adjacent Vertices: Two vertices connected by an edge

Neighbor: A vertex directly connected to another vertex

Subgraph: A graph formed from a subset of vertices and edges from a larger graph

Connected Component: A maximal set of vertices where each pair has a connecting path

Root: The top vertex in a tree structure

Leaf: A vertex with degree 1 in a tree (no children)

==References==

[[Category: Mathematics]]

Graph Theory

2026-02-25T19:28:04Z

Tom: /* Taxonomy */

==Meme==
[[Graph Theory]]
==Context==
[[Trust Graph]]
==Taxonomy==
# By Structure:
## Undirected Graph: Edges have no direction; connections are bidirectional
## Directed Graph (Digraph): Edges have direction; connections flow one way
## Weighted Graph: Edges have assigned values or costs
## Unweighted Graph: All edges are equal; no values assigned
# By Connectivity:
## Connected Graph: Path exists between every pair of vertices
## Disconnected Graph: Some vertices have no connecting path
## Complete Graph: Every vertex connects to every other vertex
## Tree: Connected graph with no cycles
# By Cycles:
## Acyclic Graph: Contains no cycles or loops
## Cyclic Graph: Contains at least one cycle
## DAG (Directed Acyclic Graph): Directed graph with no cycles
# By Representation:
## Simple Graph: No self-loops or multiple edges between vertices
## Multigraph: Multiple edges allowed between vertices
## Pseudograph: Self-loops and multiple edges allowed
# By Application:
## Social Network Graph: Represents relationships between people
## Knowledge Graph: Represents entities and their relationships
## Flow Network: Models flow of resources through a network
## Dependency Graph: Shows dependencies between elements
## [[Trust Graph]]: Allows evaluation or attestation of Entities (Nodes)

Node (Vertex): A fundamental point or entity in a graph that represents an object or data point

Edge (Link/Arc): A connection between two vertices that represents a relationship or path

Degree: The number of edges connected to a vertex

In-degree: Number of incoming edges (directed graphs)
* Out-degree: Number of outgoing edges (directed graphs)
* Weight: A numerical value assigned to an edge representing cost, distance, or strength

Path: A sequence of vertices connected by edges

Cycle: A path that starts and ends at the same vertex

Loop (Self-loop): An edge that connects a vertex to itself

Adjacent Vertices: Two vertices connected by an edge

Neighbor: A vertex directly connected to another vertex

Subgraph: A graph formed from a subset of vertices and edges from a larger graph

Connected Component: A maximal set of vertices where each pair has a connecting path

Root: The top vertex in a tree structure

Leaf: A vertex with degree 1 in a tree (no children)

==References==

[[Category: Mathematics]]

Graph Theory

2026-02-25T19:27:17Z

Tom: /* References */

==Meme==
[[Graph Theory]]
==Context==
[[Trust Graph]]
==Taxonomy==
# By Structure:
## Undirected Graph: Edges have no direction; connections are bidirectional
## Directed Graph (Digraph): Edges have direction; connections flow one way
## Weighted Graph: Edges have assigned values or costs
## Unweighted Graph: All edges are equal; no values assigned
# By Connectivity:
## Connected Graph: Path exists between every pair of vertices
## Disconnected Graph: Some vertices have no connecting path
## Complete Graph: Every vertex connects to every other vertex
## Tree: Connected graph with no cycles
# By Cycles:
## Acyclic Graph: Contains no cycles or loops
## Cyclic Graph: Contains at least one cycle
## DAG (Directed Acyclic Graph): Directed graph with no cycles
# By Representation:
## Simple Graph: No self-loops or multiple edges between vertices
## Multigraph: Multiple edges allowed between vertices
## Pseudograph: Self-loops and multiple edges allowed
# By Application:
## Social Network Graph: Represents relationships between people
## Knowledge Graph: Represents entities and their relationships
## Flow Network: Models flow of resources through a network
## Dependency Graph: Shows dependencies between elements
## [[Trust Graph]]: Allows evaluation or attestation of Entities (Nodes)

Vertex (Node): A fundamental point or entity in a graph that represents an object or data point

Edge (Link/Arc): A connection between two vertices that represents a relationship or path

Degree: The number of edges connected to a vertex

In-degree: Number of incoming edges (directed graphs)
* Out-degree: Number of outgoing edges (directed graphs)
* Weight: A numerical value assigned to an edge representing cost, distance, or strength

Path: A sequence of vertices connected by edges

Cycle: A path that starts and ends at the same vertex

Loop (Self-loop): An edge that connects a vertex to itself

Adjacent Vertices: Two vertices connected by an edge

Neighbor: A vertex directly connected to another vertex

Subgraph: A graph formed from a subset of vertices and edges from a larger graph

Connected Component: A maximal set of vertices where each pair has a connecting path

Root: The top vertex in a tree structure

Leaf: A vertex with degree 1 in a tree (no children)
==References==

[[Category: Mathematics]]