Talk:Artificial intelligence

On going issues

Goals

I think a high level listing of AI's goals (from which more specific Problems inherit) is needed; for instance "AI attempts to achieve one or more of: 1) mimicking living structure and/or internal processes, 2) replacing living thing's external function, using a different internal implementation, 3) ..." At one point in the past, I had 3 or 4 such disjoint goals stated to me by someone expert in AI. I am not, however. DouglasHeld (talk) 00:11, 26 April 2011 (UTC)

We'd need a reliable source for this, such as a major AI textbook. ---- CharlesGillingham (talk) 16:22, 26 April 2011 (UTC)

Length

I argue that this is WP:Summary article of a large field, and that therefor it is okay that it runs a little long. Currently, the article text is at around ten pages, but the article is not 100% complete and needs more illustrations. ---- CharlesGillingham (talk) 18:29, 2 November 2010 (UTC)

Todo: Illustration

The article needs a lead illustration and could use more illustrations throughout. ---- CharlesGillingham (talk) 18:29, 2 November 2010 (UTC)

Thanks to User:pgr94, the article is 70% illustrated. Almost there. ---- CharlesGillingham (talk) 00:03, 16 June 2011 (UTC)

Main illustration doesn't provide an actual example of an Artificial Intelligence, just a robot capable of mimicking human actions in a certain area (Namely, sport) — Preceding unsigned comment added by 86.163.226.52 (talk) 15:37, 4 August 2011 (UTC)

Todo: Applications

The "applications" section does not give a comprehensive overview of the subject. ---- CharlesGillingham (talk) 18:29, 2 November 2010 (UTC)

Todo: Topics covered by major textbooks, but not this article

I can't decide if these are worth describing (in just a couple of sentences) or not. ---- CharlesGillingham (talk) 18:29, 2 November 2010 (UTC)

1. Could use a tiny section on symbolic learning methods, such as explanation based learning, relevance based learning, inductive logic programming, case based reasoning.
2. Could use a tiny section on knowledge representation tools, like semantic nets, frames, scripts etc.
3. Control theory could use a little filling out with other tools used for robotics.
4. Should mention Constraint satisfaction. (Under search). Discussion below, at Talk:Artificial intelligence/Archive 4#Constraint programming.
5. Should mention the Frame problem in a footnote at least. ---- CharlesGillingham (talk) 19:52, 3 February 2011 (UTC)

Todo: redlinks and tags

1. Where can we link Belief calculus? Does this include Dempster-Shafer theory (according to R&N)? I think that's more or less deprecated. Does R&N include expectation-maximization algorithm as a kind of belief calculus? I don't think so. Where is this in Wikipedia?
2. There are still several topics with no source: Subjective logic, Game AI, etc. All are tagged in the article. ---- CharlesGillingham (talk) 19:59, 3 February 2011 (UTC)

Recent copyedits

A series of recent changes which bring this article in line with the layout more typically used in articles was rolled back here. The most significant of these changes were the reintroduction of large HTML comments to separate different content areas, and the reintroduction of a great deal of superfluous markup in the references section. The former is unnecessary: we do not habitually use comments in this manner in other articles, and especially in the article introduction it should be obvious that the separate paragraphs refer to separate concepts simply by looking at them. As for the references section, the markup used is simply highly idiosyncratic and makes the actual footnote text nearly unreadable (and that's on a desktop browser: I shudder to think how legible it is on a mobile device). In the interests of collaborative editing it would be better for this page to more closely follow the practices of other articles in these regards.

Furthermore, the above idiosyncracies do not apply solely to the article. It is not clear why a todo list is being maintained as a page section rather than being moved to /to do, as is common practice. If there are no objections I'd like to move it there, as ideally comments on the main talk page should not be repeatedly re-edited.

Chris Cunningham (user:thumperward) - talk 10:00, 26 June 2011 (UTC)

Let's just take it point by point.

I like the Todo list using {{Todo}}; let's implement that.

Next, the references section. I take it you claim that the your version is easier to edit than the current version? I've reprinted them in the following section (your version is first, the current version is second). I am not convinced. It is difficult to find an individual footnote in your version. (By the way, this reference section conforms to the standards supported by WP:CITE. These are topical bundled footnotes as per WP:CITEBUNDLE, using shortened footnotes for references used multiple times (see WP:CITESHORT). We've used list-defined references (see WP:LDR) because the sheer volume of reference makes it too difficult to edit the text. See our discussion above.)

Finally the hidden topic descriptions for paragraphs. (Such as <!-- AI at MIT -->) In my experience, it is very common for new contributions to be off of the topic of a paragraph or section, especially in an article such as this. These hidden topic descriptions are a way to try to draw the contributor's attention to the topic of the paragraph and help them to find the right place to add a sentence or two. This is an innovation, and, like all innovations, it is currently idiosyncratic. I think that one could at least agree that these hidden comments cause no harm. ---- CharlesGillingham (talk) 11:06, 26 June 2011 (UTC)

From the top:

1. I'll move the todo later. Thanks.
2. The problem is the sheer volume of footnotes, together with the provision of an average of three references for almost every assertion made in the article. Fully 50% of the entire page size consists of references, and using the current citation style nearly 75% of the article by length is footnotes. The footnotes also intersperse citations with commentary (such as the current footnote 7) or quotations (such as the current footnote 80), and many of the footnotes are not citations at all but rather general pointers to useful reading (such as the current footnote 12). What's more, the vast majority of the footnotes refer to the five textbooks. It is unclear that there is a benefit from such deep referencing of the same sources over and over again, especially when an additional "Other sources" section with no footnotes is provided as well. But quite aside from this, the use of nested {{refbegin}} templates to make the cited text tiny should certainly be stopped even if the current footnotes are kept.
3. The use of hidden comments simply exacerbates the problem with the relative code-to-content ratio in the article. Were the (IMO) over-the-top use of footnotes curtailed, there would be less need for comments to navigate them by.

Chris Cunningham (user:thumperward) - talk 11:55, 26 June 2011 (UTC)

Validation over Reach of A.I. Market in the 1980s

I would only like to know the Market Study, the Specialized Article or any Serious Source for the Affirmation "By 1985 the market for AI had reached over a billion dollars", because I have been searching for It since Long Time Ago and cannot find Anything. AFGV 03 March 2012 (UTC) — Preceding unsigned comment added by 181.135.62.175 (talk)

Content of the Artificial Intelligence article

Hi all,

Excuse me I'm french, my english is not good. In my opinion, this article teaches absolutely nothing about what is artificial intelligence and it is misleading. It is too long, it's a catch-all, all is mixed. The lie is there : this article speaks constantly of the many successes of AI, when in fact its results are a failure until today. If AI was a succes "In the 1990s and early 21st century", all IT today would be AI. This is not the case ! AI is a broken promise, an ambition and its few successes are ignored (see Jean-Philippe de Lespinay). "In the early 1980s, AI research was revived by the commercial success of expert systems" ! "By 1985 the market for AI had reached over a billion dollars" ! The impressing "Japan's fifth generation computer project" (was in fact a bitter defeat for the Japanese, for example in front of Turbo Prolog/Borland ...on a simple PC ! ), "In the 1990s and early 21st century, AI achieved its greatest successes" !!! Horror! This is journalism, not economic facts.

The history of AI did not go like that. You might interview Jean-Philippe de Lespinay, who is still alive and lived this history from 1982 until today. His job was AI developments non stop from 1982. You have here two samples of "good" articles about AI and expert systems (unfortunately in French) : Larousse encyclopedia: Definition of Artificial Intelligence and Larousse encyclopedia: Definition of the expert system.

Best regards and sorry for the criticism

Pat grenier (talk) 17:48, 20 August 2011 (UTC)

I think you missed the sentences where the article specifically calls early AI a failure: "They had failed to recognize the difficulties of the problems they faced". This is harsh enough, I think. The article also mentions that AI failed a second time in the 1980s and early 90s: it says "AI once again fell into disrepute". This is referring directly to expert systems and the Fifth Generation Project. Again, I think this is about as harsh as an encyclopedia needs to be. See also the History of AI article, which has more detail about these failures. (The "history" section in this article is really a summary of that article.)

AI's success in recent decades is in fields like machine learning, pattern recognition and robotics, where the "statistical" approach has provided the technology industry with a lot of successful applications. This is well documented in the sources. Again, see history of AI for more detail. Note that the "statistical approach" has relatively modest goals; they provide specific solutions to specific problems. No one is trying to build a machine that can do "all IT today". That's not what AI is any more. ---- CharlesGillingham (talk) 07:13, 26 August 2011 (UTC)

Here are a few more of the Larousse article's claims:

• the "statistical" approach that solves small applications is not AI and so a lot of software in industry claims to be AI when it is not.
• Deep Blue was not AI, there was no reasoning in it at all. ('Pas le moindre raisonnement là-dedans.') While a human a simulates plies, compares them, develops a strategy, reasons, and in a glance recognises the value or danger of a position, computer chess programs just calculate.
• AI is the automation of human reasoning. (’Intelligence Artificielle, c’est d’abord l’automatisation du raisonnement humain.')

For brevity, I've lost some of the nuance, but these claims stuck out the most. pgr94 (talk) 11:04, 26 August 2011 (UTC)

Another interesting paragraph (thanks to google translate, plus a few minor fixes, my emphasis):

It was in France (Université Paris VI) that the expert system became a powerful tool, demonstrating that AI was not a chimera, with the appearance of an engine with the logic called "first order." This strange name describes our daily logic, which allows us to explain to others or to infer new facts. It just needed programming, which was not difficult. Only French researchers have done this (Pandora expert system) ... then have forgotten it, not realizing the scope of their discovery. Yet it was the first to show the aspects "mental" claimed by Minsky: real thinking skills, ability to solve problems, to decide, learn and communicate ("conversational") based on reasoning, the ability to explain what it "thinks" and even the ability to evaluate knowledge by the detection of contradictions (Minsky's "critical thinking"). Its most famous example was the expert system generator "Intelligence Service", derived from a commercial product marketed by Pandora Sociétés GSI-TECS and Arcane, who sold hundreds of copies. The expert system of first order is the first tool and fully operational techniques from AI. At the same time, it is the first to have demonstrated the feasibility of what some call the "strong" AI.

pgr94 (talk) 11:23, 26 August 2011 (UTC)

pgr94, pardon my english, please, I too am French. I don't understand if you suggest to keep the whole article or if you accept its modification. The same article doesn't make the apology of AI citing its many successes if AI is an echec, even acknowledging its failures. The reader should understand what it is and if it works. With this article, it is impossible. The certainty that the AI article should show it would be AI ​​was not successful, until now. It has had no operational success, at least in the United States (Mycin included).

I cite you: "AI's success in recent decades is in fields like machine learning, pattern recognition and robotics, where the "statistical" approach has provided the technology industry with a lot of successful applications". But the technics used in that domains are classic technics ! They are programmed by developers with classic languages (C++, Pascal, etc.) Where is AI ? In the newspaper only. Wikipedia repeats what the press says without any control. Newspapers need sensationalism and "researchers" need publicity. And Wikipedia repeats the two. Robotics is no AI. You can put AI inside but nobody did it (to my knowledge). You can make pattern recognition and machien learning with clssic technics. How to know if it is AI or not ?

The problem of AI is that everyone claims to do it. The article should begin by saying what "intelligence" is and what "artificial" means. It must begin by the historical definition of Mc Carthy (google translation of a french citation translated of an english citation !): "The construction of computer programs that engage in tasks that are, for now, more satisfactorily accomplished by human beings because they require high-level mental processes such as perceptual learning, organization memory and critical thinking".

The article should tell there is two AI: the "true" AI which is a technologoical breakthrough and the "false" AI which is an ambition (= a publicity) put in the programs with classic technics.

The true AI is based on real human intelligence: it must at least reason, like all animals can do. Jean-Philippe de Lespinay (talk) 10:05, 29 August 2011 (UTC)

— Preceding unsigned comment added by Jean-Philippe de Lespinay (talk • contribs) 09:01, 29 August 2011 (UTC)

Perhaps the article should make it more clear that people disagree about what AI is. These questions are addressed, very briefly, at the top of the section on "approaches". The definition of AI used in this article comes from the most widely-used textbook, Russell & Norvig. I don't think there is a more reliable source. (Note that, here at Wikipedia, we don't attempt to write the "truth"; we only report what WP:Reliable sources claim, regardless of whether we personally believe it is "true" or not.) ---- CharlesGillingham (talk) 21:52, 29 August 2011 (UTC)

I should also note that all of the points that are made above are covered by either this article or History of AI, with one exception: neither article claims that a particular approach was the only "true" AI. ---- CharlesGillingham (talk) 22:04, 29 August 2011 (UTC)

I added a sentence criticizing the statistical school of AI along the lines you suggest. Search for a heading marked Statistical and read the last sentence. Do you think you could find a citation for this? ---- CharlesGillingham (talk) 22:10, 29 August 2011 (UTC)

Charles, glad to see you are really interested in debating about this article. Yes, "the article should make it more clear that people disagree about what AI is", I agree. And that from the beginning of the article. You say Russell & Norvig is the most widely-used textbook about AI, but for whom ? For neophytes or for AI specialists ? Surely not for specialits. In my opinion, one of the most reliable sources it's me (I beg your pardon...): I'm an expert in AI, I managed an AI company during 16 years (1986-2002), I produce AI during 25 years, I invented in that domain, I gave reality to the concepts of Minsky and others, I sold my AI to big companies and administrations. I published various articles about AI including AI and expert system definitions in an encyclopedia (Larousse). My suggestion is to redo the article, in order it will be short and clear. What do you think?

"Critiques argue that these techniques are too focussed on particular problems and have failed to address the long term goal of general intelligence" You want a citation. But you can repeat that about each part of AI : fuzzy loggic, computational intelligence (a joke !), neural networks, machine learning, natural language, intelligent agents, intelligent chatterbots, pattern recognition, bayesian networks and unfortunatly too often expert systems. You understand surely that using the term "artificial intelligence" about a software sells it ! It's business hoax ! And who is in charge to verify it ? Nobody... It's why Wikipedia must clarify what is AI in the reality (technology) and in the profanes world. Jean-Philippe de Lespinay (talk) 14:34, 30 August 2011 (UTC)

I agree with you, in general. I worked in AI from 1988 to 1994 myself; we're from a similar generation and I share your skepticism that current research is on the right track. I think you would agree with me that, at some point, the field must revisit logic and knowledge because these are poorly handled in the current statistical framework. Although we may disagree on the precise details here, I think our general outlook is the same.

However, as I said above, my opinion doesn't really matter. This article must reflect the current consensus, regardless of what you or I think. For better or worse, introductory college courses provide a simple, measurable and objective way to determine what the current consensus is. The plurality of introductory college courses use Russell & Norvig.

Thus the article reports that, according to the current consensus, AI is succeeding; we are in an "AI summer". AI research (and statistical AI in particular) have developed a huge number of successful applications. These handle tasks that would otherwise require human intelligence, and therefor they satisfy McCarthy's definition of AI that you quoted above. They do not fit Minsky's definition of AI, or yours, or mine. However, the article must report that these applications exist and that they are successful. We must also report that researchers who have created these applications call their field "AI" (for the most part), even when their techniques are drawn entirely from other fields. ---- CharlesGillingham (talk) 22:33, 30 August 2011 (UTC)

OK, we are both old men with lived experience of AI. We are the memory of the 20th century. Then it's OUR responsibility to inform our contemporaries. We agree on what AI is and what it is not. We should write the same article ! But, there would be a particular vision of the encyclopedia that would require to hide the truth to be "consensual" ? I can't believe it. Wikipedia should reflect our debate. We must inform in Wikipedia that AI ​​has achieved all the goals defined by Minsky 40 years ago : perceptual learning, organization of memory and critical reasoning. We must describe the fantastic concept of the expert system (a tool designed to reason like humans), we must inform about founding experience of Mycin and its shortcomings of the time ("strong" AI ?). Are you agree ? And OK to inform that there is another vision of AI in the business world, more promotional and much less demanding (weak AI). We must also put AI ​​in the modern context where it is really called for and needed: video games. The requirement of the market is so strong, the market volume is growing so much, pragmatism is so important that this is probably there that strong AI will finally appear before people.

Charles, I don't want to continue this (interesting) discussion and mobilize your attention only for the fun. If you think the article can't be changed, let me know and I'll stop there. Jean-Philippe de Lespinay (talk) 09:54, 31 August 2011 (UTC)

I think you'll find it very hard to change Wikipedia. It should be much easier to publish these opinions in respected AI journals. Once published these perspectives can appear in Wikipedia. That seems like the path of least resistance to me. Also please see the Expert system article where expert systems are already covered. Perhaps you could help improve this article? pgr94 (talk) 10:45, 31 August 2011 (UTC)

Thank you for that suggestion, Pgr94. If you want a "respected AI journal", I almost have it but in french :

1) In Science et Vie article I expose in may 1991 AI don't produce anything usefull, researchers do not leave their office and doing research in AI completly disconnected from the realities and needs of businesses, with useless logic (epistemic, modal, fuzzy, "n" order logic, temporal, etc.). I noted they they didn't even see the benefit of the invention of the logic of order 0, owned by all human, nor the Pandora prototype, a french university project of expert system (1987-88) that worked by reasoning with the Zeroth-order logic. "Despite all its troubles AI has managed to give birth, dying, zeroth order expert systems, the only operational IA technique (at the time and today)"

2) In Automates Intelligents, a scientific website, I present in 2008 "reasoning AI", repeating that AI ​​is in full decay and produced one interesting thing : the expert system.

It is surely easy to find english articles in respected journals telling the business world is desappointed by IA results and AI is an illusion. With that kind of articles, it is possible to rebuild wikipedia article about AI and say there is two AI and no more. The strong one and the weak one.

About your suggestion to see expert system article, which person of Wikipedia is in charge of that article please ?Jean-Philippe de Lespinay (talk) 16:06, 31 August 2011 (UTC)

It seems Science et Vie is popular science magazine, not a peer-reviewed scientific journal. For more information about reliable sources, please see WP:RS, and in this case WP:SCHOLARSHIP. Regarding who is charge of an article, Wikipedia is a collaborative encyclopaedia that anyone can edit providing they observe the fundamental principles: WP:FIVEPILLARS. Hope that helps. pgr94 (talk) 08:51, 1 September 2011 (UTC)

Thank you pgr94.

"It seems Science et Vie is popular science magazine, not a peer-reviewed scientific journal". Yes, that's right. In France you can't publish in "official" scientific journal if you are a private researcher ! The publications are reserved for academics, who are public servants, who are representatives of the french State. Or by companies who have done their R & D thanks to university research. You must understand that when you think french research. The proof is the hundreds of articles published about my inventions and their installations in companies and ... zero article in the official journals. You can't imagine the State repression I suffered from the day I was published in Science and Vie, the most widely read scientific journals in Europe. It would take a whole book to tell it. Jean-Philippe de Lespinay (talk) 10:36, 1 September 2011 (UTC)

Expert systems: eyes please

Recently there have been signicant changes to the expert systems article. I wonder if others would care to give their opinion. pgr94 (talk) 18:34, 30 September 2011 (UTC)

Looks very fishy. I've made a request at Wikipedia:Sockpuppet investigations/Pat grenier. —Ruud 21:26, 4 October 2011 (UTC)

This is a difficult problem and should be handled carefully. What's most important here is that we WP:assume good faith and just try to clean up the article. I think what we have is someone who is an expert in the field, who (1) plays loose with WP:VER because he tends to use citations like academic citations, rather than as verifications. And (2) might be WP:POV-pushing. This sort of thing has happened before ...the Carl Hewitt affair comes to mind. ---- CharlesGillingham (talk) 08:34, 7 October 2011 (UTC)

Hewitt was a computer scientist with an established track record. The person here doesn't come from academia (or at least hasn't managed to leave any trace of his existence behind in academic journals or proceedings). The only thing I could find are the two articles in Larousse ([1] [2]). Apparently they allow anyone to contribute, though. —Ruud 12:04, 10 October 2011 (UTC)

Il is true : everyone can create a compte contributeur and then write articles. --Rigoureux (talk) 14:07, 10 October 2011 (UTC)

Logic

Logic section presently has sentence beginning The study of logic led directly to the invention of, which is flat wrong. The study of logic had to detour through George Boole and An Investigation of the Laws of Thought (1854), on Which are Founded the Mathematical Theories of Logic and Probabilities, before 0 + 1 could equal anything but 1, and before there were any useful means of analyzing artificial thought or artificial intelligence. The omission is illogical. He should certainly get prominent mention, and editors might even consider honorable mention of a Logic Named Joe.--Pawyilee (talk) 05:29, 30 October 2011 (UTC)

There are many centuries of people who are equally important, including those who came before Boole (Aristotle, Euclid, al-Khwārizmī, Liebniz) and those who came after (Frege, Russell, Church, Post, Godel, Turing, Von Neumann). The "programmable digital computer" is the end result of a train of thought that includes all these people. The article mentions only Turing, because it was Turing who developed a "mechanical" metaphor to describe all of mathematical logic. The sentence that says "the study of logic..." is referring to the fact that all of these people where trying to improve and/or understand mathematical logic. That's what they were "studying" when they stumbled onto the idea of a computer. ---- CharlesGillingham (talk) 18:57, 30 October 2011 (UTC)

In doing so, they stumbled onto a 17th-century word for teller. --Pawyilee (talk) 04:02, 5 November 2011 (UTC)

Fixing citations

I have fixed some cites which were not linking to the citations/works/books to which they were apparantly meant to link [3]. I have also deleted two cites because there seemed to be no cited book (in the article) to which they could be linked. I would request that some regular editors to this article may review my edits. Thanks.MW ^ℳ 16:01, 12 November 2011 (UTC)

I would like two edits in particular to be reviewed. In those edits, I had deleted citations. This could mean that some of the material in the article is now unsupported by proper citations.

• The first is this edit, [4] which should relate to the lines Many futurists believe that artificial intelligence will ultimately transcend the limits of progress. Ray Kurzweil has used Moore's law (which describes the relentless exponential improvement in digital technology) to calculate that desktop computers will have the same processing power as human brains by the year 2029. He also predicts that by 2045 artificial intelligence will reach a point where it is able to improve itself at a rate that far exceeds anything conceivable in the past, a scenario that science fiction writer Vernor Vinge named the "singularity".[165] in the article.
• The second edit, [5] should be related to this Roger Penrose is among those who claim that Gödel's theorem limits what machines can do. (See The Emperor's New Mind.)[155]

If we have no citations to support what Vernor Vinge and Roger Penrose are being made to say in the article, we may need to delete the points attributed to them. That is why I suggested that my edits be reviewed. Thanks.MW ^ℳ 13:21, 13 November 2011 (UTC) Besides these two cites, I had also deleted one citation to Picard. I think that should be OK.MW ^ℳ 13:31, 13 November 2011 (UTC)

I'm adding the Penrose and Vinge citations. ---- CharlesGillingham (talk) 10:02, 14 November 2011 (UTC)

I would be happy if those citations can be added. My objection was that we seem to be attributing some views to Penrose and Vinge through some books / articles etc. written by Penrose and Vinge. Only that our article does not seem to name/identify any book or article or paper or whatever written by Penrose and Vinge.MW ^ℳ 10:11, 14 November 2011 (UTC) I see that you have added the relevant cites now. Thanks.MW ^ℳ 10:29, 14 November 2011 (UTC)