Talk:History of artificial intelligence

GA Reassessment

History of artificial intelligence

The following discussion is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

---

Article (edit | visual edit | history) · Article talk (edit | history) · Watch • Watch article reassessment page • Most recent review

Result: Consensus to delist. ~~ AirshipJungleman29 (talk) 08:28, 13 July 2023 (UTC)

The talk page of this 2008 listing was tagged by SandyGeorgia as requiring a GAR; I must agree. The article has not been updated to the sufficient standard after 2010; this is especially egregious considering the massive leaps in AI over the last decade.

Thus, I'll tag it as needing an {{update}}, and nominate this for delisting as failing GA criterion 3. ~~ AirshipJungleman29 (talk) 18:50, 4 July 2023 (UTC)

I agree that this article needs huge amounts of work and updating to be at standard. Should be delisted unless someone takes that on. SandyGeorgia (Talk) 01:21, 5 July 2023 (UTC)

agree, should be delisted. Section for 2011 is really outdated and needs a huge amount of work Artem.G (talk) 06:21, 7 July 2023 (UTC)

Delist. Needs significant effort. If anyone steps forward to work on this article, please ping me. BennyOnTheLoose (talk) 13:28, 10 July 2023 (UTC)

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

19th century fiction

Shouldn't E. T. A. Hoffman's stories ( The Sandman (1816) and Automata (1814) ) be mentioned? Kdammers (talk) 21:08, 30 October 2023 (UTC)

In my opinion, this article has too many fictional and mythological precursors already. CharlesTGillingham (talk) 08:44, 31 July 2024 (UTC)

Cut for brevity / lack of notability

None of the major overviews (Russell & Norvig, McCorduck, Crevier, Nilsson, Newquist) mention WABOT, as far as I know. ---- CharlesTGillingham (talk) 19:56, 3 August 2024 (UTC)

====Automata====

In Japan, Waseda University initiated the WABOT project in 1967, and in 1972 completed the WABOT-1, the world's first full-scale "intelligent" humanoid robot, or android. Its limb control system allowed it to walk with the lower limbs, and to grip and transport objects with hands, using tactile sensors. Its vision system allowed it to measure distances and directions to objects using external receptors, artificial eyes and ears. Its conversation system allowed it to communicate with a person in Japanese, with an artificial mouth.

CharlesTGillingham (talk) 19:56, 3 August 2024 (UTC)

Cut this as well for brevity. I'm under the impression that specialized hardware did not have last influence and wasn't widely used. Most work was on digital computers and the most influential work of the time (1980s) was theoretical.

The development of metal–oxide–semiconductor (MOS) very-large-scale integration (VLSI), in the form of complementary MOS (CMOS) technology, enabled the development of practical artificial neural network technology in the 1980s.

A landmark publication in the field was the 1989 book Analog VLSI Implementation of Neural Systems by Carver A. Mead and Mohammed Ismail.

References

---- CharlesTGillingham (talk) 04:31, 4 August 2024 (UTC)

Protein structure prediction

There is one domain where Deep learning (not yet called that) was successful as early as the end of 1980s, the prediction of protein structures. People like Terry Sejnowski started to use neural net to predict secondary structures

N Qian, TJ Sejnowski (1988) Predicting the secondary structure of globular proteins using neural network models. Journal of molecular biology, 202 (4): 865-884 (cited 1700 times)

And in 1993, Rost and Sander proposed a cascading neural net structure, PHD, that basically killed the field by reaching theoretical maximum accuracy.

Rost, Burkhard, and Chris Sander (1993) Improved prediction of protein secondary structure by use of sequence profiles and neural networks. Proceedings of the National Academy of Sciences 90.16: 7558-7562. (cited 3900 times)

(well, the absolute best was actually PsiPred, an improvement by David Jones a bit later, using profile matrices rather than multiple sequence alignments

McGuffin, Liam J., Kevin Bryson, and David T. Jones (2000) The PSIPRED protein structure prediction server." Bioinformatics 16.4: 404-405. (cited > 4000 times)).

Ahaemd 39.56.202.138 (talk) 16:11, 30 January 2025 (UTC)

DeepSeek has not been mentioned

Deepseek is a very important milestone in AI development. It should be mentioned in the article. Robot8in (talk) 13:19, 22 February 2025 (UTC)

Expert systems are not AI

In the section "Boom (1980–1987)" it says 'a form of AI program called "expert systems"' Mousing over it says "an expert system is a computer system emulating the decision-making ability of a human expert."

They're just a convoluted load of explicitly coded if/then/else statements based on human rules & heuristics, like "don't squat when you're wearing spurs" and "never start a land war in Asia".

In short, they're no more intelligent than a text editor and they're barely artificial. ~2025-32326-81 (talk) 15:59, 9 November 2025 (UTC)

Maybe from a 21st century point of view. However, in the 1980s, they were the primary focus of the academic and industrial research program called "Artificial Intelligence", which is the subject of this article. ---- CharlesTGillingham (talk) 07:09, 13 March 2026 (UTC)

Trimming history sections

This is regarding these two edits. My intention was to trim some of the redundancy and tangents from the article, for brevity and focus. Much of this was originally added in 2020 by a student editor, and to be blunt, it reads like it was intended to meet a quota for a grade. There were also too many MOS:OVERLINKs and editorializing asides. The article should mainly summarize secondary sources which directly connect primary sources to this topic itself, otherwise picking examples of magical life forms from throughout history is a form of WP:OR. I would also be cautious of citing David Berlinski. He is prolific and florid, so any examples he cites may not be due weight, and there are also are WP:FRINGE issues. Grayfell (talk) 03:36, 17 November 2025 (UTC)

I agree that all the extra information was not relevant to historical AI or modern AI. I cut it all, except the names (the reader can follow the link if they're interested). Added a sentence to explain why these stories and myths might still have some relevance. ---- CharlesTGillingham (talk) 09:03, 13 March 2026 (UTC)

I'm wondering a bit about the neurology that you cut -- clearly it is the inspiration for "neural networks" and AI directly borrowed Hebbian learning and other things.

AI also borrowed from psychology (e.g., reinforcement learning), statistics & mathematical optimization (e.g. stochastic gradient descent), probability theory (e.g. Bayesian inference), economics (e.g. decision theory), control theory / cybernetics, (e.g. "objective functions") operations research (e.g. means-ends analysis) and so on.

Russell & Norvig cover each of these in their history section with about a page each -- they think this is a part of the "history of AI". They call them "foundations" of AI.

Maybe we could cover them really rapidly -- a sentence for each field. R&N as the source. Does that make sense to you? ---- CharlesTGillingham (talk) 09:03, 13 March 2026 (UTC)

Confusion of Symbols and the Physical Symbol System Hypothesis

In the section about Newell and Simon there begins (and persists) a common confusion between "symbolic computing" and what Simon and Newell called the "Physical Symbol System Hypothesis" (PSSH). This is an important thing to get straight. There is a long WP entry on PSSH, although I think it misses some important nuance that the PSSH may yet be correct. (We do not know why LLMs are so performant in some areas, and it may well be that they have developed exactly what Newell and Simon referred to as symbols.) But what Newell and Simon were NOT saying is that people had a symbol table and lists in their head in the same sense as Lisp does that was a convenient computational simplification, and turned out to be brittle, as many predicted at the time. (That said, note that there's a whole lotta code in the world not built on ANNs that actually does stuff that you and everyone else depends upon, so let's not be so fast to diss anything that's not an ANN.) I think that the right thing to do here is slightly revise the Newell and Simon part, and point to the PSSH page, and then I'll go over there and argue with those folks about whether PSSH == ~Lisp (it doesn't, and never did!) Jshrager (talk) 14:41, 13 March 2026 (UTC)

The PSSH explicitly says that physical symbol systems are necessary for "general intelligent action". This implies that anything that is "intelligent" (like us) must be using "physical symbols" manipulated by some "system". So they are saying that we do manipulate a Lisp-like list in our heads -- e.g., when we are planning, we iterate through a list of possible actions we could take. When we solve a problem, we consciously consider each of a list of possible partial-solutions, and so on.

Psychologists like Daniel Kahnemann proved that people almost never use step-by-step reasoning. Most of the time, we just intuitively guess at the right answer -- it just "pops into our heads" (Kahnemann calls this "system 1" problem-solving). LLMs work similarly; they're making contextually circumscribed stochastic predictions, which is basically the same thing as a guess.

So both of these lines of evidence (psychology, LLMs) show that the PSSH is dead wrong. Step-by-step reasoning isn't necessary or sufficient for intelligent behavior.

This does not mean that step-by-step symbolic reasoning isn't useful -- on the contrary, we use it to carefully check that we are right about our guesses. This is something that LLMs can't do very well (yet), but other deep learning systems (such as Alpha Zero) use it. ---- CharlesTGillingham (talk) 01:24, 14 March 2026 (UTC)

From the October 7, 1974 interview with Pamela McCorduck talking to Allen Newell at Carnegie-iMellon University (Newell_box00088_fld06028_doc0003.pdf in the McCorduck archives at CMU): "So that what one has here is a proposal, a hypothesis, that what really is going on in symbolization inside the human is just the fact that there are patterns that occur in expressions[note] which when the system encounters those patterns, it's able to use that pattern to gain access to all the other information, and that is in fact what meaning is all about." [Note] Expression can jut be larger patterns. Sounds just as likely to be pattens of neural or ANN gunk as s-expressions. Jshrager (talk) 02:01, 14 March 2026 (UTC)

Here's the argument spelled out by a former student of Newell's: https://arxiv.org/abs/2306.13150 (Published at the the 16th Annual AGI Conference, 2023) Jshrager (talk) 18:07, 14 March 2026 (UTC)

Nice source (Newell & McCorduck). I think the key here is the way they are using word "expressions" -- I think it's pretty clear (given the literature of the time) that they are talking about s-expressions.

Speaking directly to your point, I like the way Nils Nilsson makes a distinction between "signals" and "symbols". An ANN has digitized signals, which is not the same thing as s-expressions. If we construe the PSSH as talking about digitized signals, then the PSSH is trivially true, because literally anything can be digitally simulated,[1] In order to be interesting, the PSSH has to be be talking about s-expressions of discrete meaningful symbols, not just any digital signal. ---- CharlesTGillingham (talk) 05:12, 15 March 2026 (UTC)

Okay. Peace. At least can be put a nod to the PSSH here and a pointer to the WP article on it, and then take this argument over there? Jshrager (talk) 05:48, 15 March 2026 (UTC)

[1] I should have said, "the sufficient side of the PSSH is trivially true". And I should also have said, "anything can be digitally simulated to given degree of accuracy, given representative data, the right program, enough memory and enough time. This is just the real-world corollary to the Church-Turing Thesis.". ---- CharlesTGillingham (talk) 05:12, 15 March 2026 (UTC)

Semi-protected edit request on 10 April 2026

Following the paragraph: "Neural networks, along with several other similar models, received widespread attention after the 1986 publication of the Parallel Distributed Processing, a two-volume collection of papers edited by Rumelhart and psychologist James McClelland. The new field was christened "connectionism" and there was a considerable debate between advocates of symbolic AI and the "connectionists".[112] Hinton called symbols the "luminous aether of AI"―that is, an unworkable and misleading model of intelligence.[112] This was a direct attack on the principles that inspired the cognitive revolution." add the following with references:

"Nevertheless, tools like Expert4 (published by Elsevier Biosoft in 1984 (refs 1,2)), based on psychological theories of category and concept formation (ref 3), demonstrated that inference based on measures of similarity provided computationally tractable alternative to other approaches (ref 4)."

1 - Bundy, A. and Wallen, L. (eds.) (1984), Catalogue of Artificial Intelligence Tools. Berlin: Springer-Verlag, p. 53.

2 -Anderson, A. (1985) 'Becoming an expert', New Scientist, 14 November, p. 61.

3 - Smith, E.E. and Medin, D.L. (1981) Categories and Concepts. Cambridge, MA: Harvard University Press.

4 - Spiehler, V., Spiehler, E. and Osselton, M.D. (1988) 'Application of expert systems analysis to interpretation of fatal cases involving amitriptyline', Journal of Analytical Toxicology, 12(4), pp. 216–224. Available at: https://doi.org/10.1093/jat/12.4.216 (Accessed: 19 March 2026). AI Ethics Editor (talk) 10:40, 10 April 2026 (UTC)

This work was not influential enough to be relevant in this article. ---- CharlesTGillingham (talk) 13:48, 12 April 2026 (UTC)

Expert4 introduces early practical AI tools for inductive learning and rule extraction.Also, others provide documented evidence that psychological theories of human categorization directly influenced the design of computationally tractable by AI systems. I think its helpful. Thank you. Regards, VerdictByLogic - Let's Discuss 17:34, 15 April 2026 (UTC)

 Done I have successfully added your sentence along with the citations, as they were relevant to the page. VerdictByLogic - Let's Discuss 17:32, 15 April 2026 (UTC)

I have adjusted this edit to address concerns over WP:SYNTH and WP:EDITORIALIZING. We should use reliable, secondary sources to decide whether or not some historical development was influential. I'm not saying that Expert4 wasn't influential, I'm saying we should use reliable sources to explain this instead of WP:OR on talk pages.

Please follow WP:REFPUNCT also. Grayfell (talk) 19:43, 4 June 2026 (UTC)

< Agreed. The history of AI comprises tens of thousands of research projects and literally tens of millions of academic papers. A citation to an academic paper or a news release is not enough to establish that something is notable enough to be mentioned in this article. We prefer a secondary/tertiary source that claims to provide an overview of the history of the field, such as Russell & Norvig (2021), Crevier (1993), McCorduck (2004), Newquist (1994) and Nilsson (2009) (who is under-represented at the moment). Russell and Norvig, the leading AI textbook, is an encyclopedia of AI, which mentions (at least in passing) thousands of notable research projects and theoretical advances. If it's not recent and it's not in R&N we should really think twice before covering it. ---- CharlesTGillingham (talk) 09:30, 16 July 2026 (UTC)