Difference between revisions of "Artificial Intelligence"

Revision as of 16:14, 13 February 2022

Full Title or Meme

Intelligence without human vindictiveness or human compassion.

Context

• Artificial General Intelligence redirects here. In this context Intelligence relates primarily to Identity Knowledge, but that has many general aspects.
• Some people think that Artificial Intelligence must needs be cold or sterile, but we have plenty of evidence that is not so.
  • Elisa
  • Alexa

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.
• Training an Artificial Intelligence with human behaviors will result in unacceptable behavior by the Artificial Intelligence.
  • Microsoft released Tay,^[1] a web robot (bot) to respond to tweets and chats^[2] 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; the latest from the president, Donald Trump accusing it of favoritism to leftist causes.^[3] Researcher Safiya U. Noble has written a book^[4] 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.
• Giving a AI a goal may result in unexpected and undesired solutions.
  • Bruse Schneier described^[5] 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.

Explaining AI

Most users want to know WHY a decision was make or an action was performed. The article "Solving for Why"^[6] 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.

Solutions

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 ^[7] 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^[8] who specifically talks about the need for more involvement of humanities in long term solutions.

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.^[9]

Alpha Zero

After successfully defeating the best human GO play, the Google Deep Mind project Alpha Go was generalized into a far simpler version Alpha Zero^[10] 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.

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.^[11] 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^[12] that was review by Ezra Klein.^[13]

The Smart Agent

Now that an A.I. can apply reason to problems, it should be expected that such agents on on the horizon (in 2021).

In Healthcare

• WHO Releases AI Guidelines for Health Government Technology 2021-07-12

  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.

Back to the Future

Brian Christian described "The Alignment Problem"^[14] 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^[15]. 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^[16], we can actually imagine that the brain really is just the Chinese Room that so perplex Searle (above)^[17]

References

1. ↑ Peter Lee, Learning from Tay’s introduction. (2016-05-25) https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/
2. ↑ 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/
3. ↑ Farhad Manjoo, Search Bias, Blind Spots And Google. (2018-08-31) New York Times p. B1
4. ↑ Safiya U. Noble, Algorithms of Oppression: How Search Engines Reinforce Racism. (2018-02-20) ISBN 978-1479849949
5. ↑ Bruse 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
6. ↑ Marina Kraskovsky, Solving for Why CACM 65 No 2 (2202-02) pp 11ff
7. ↑ 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/
8. ↑ Neil Savage, Neural Net Worth. CACM Vol 62 no 6 p. 12 (2019-06)
9. ↑ 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
10. ↑ 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
11. ↑ 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/
12. ↑ Sam Altman Moore's Law for Everything personal blog (2021) https://moores.samaltman.com/
13. ↑ 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
14. ↑ Brian Christian The Alignment Problem W. W. Norton (2020-10-06) ISBN 978-0-393-86833-3
15. ↑ Jean Piaget The Construction of Reality in the Child (original 1954) Ballentine ISBN 345-02338-2-165
16. ↑ Rachel Fritts, How do Babies perceive the World? MIT News July/August 2021 p 24 ff
17. ↑ James Somers, Head Space The New Yorker Dec. 6, 2021, p 30 ff.

Other Material

• See the wiki page on Intelligent Goals.
• See the wiki page on Machine Learning.