Talk:Artificial intelligence

Question mark in image

In one of the images, it asks "What is Wikipedia" without a question mark. Does this need to be changed? Asherlief4906 (talk) 21:45, 5 June 2026 (UTC)

I personally think it's not a significant problem. A lot of people omit punctuation in Google Searches so it's not too unnatural. Alenoach (talk) 23:05, 5 June 2026 (UTC)

Agree with @Alenoach. It's indicative of a 'real life' search query. SirSiliconBones (talk) 12:28, 12 June 2026 (UTC)

COI edit request: public-recognition data for AI executives (Verasight, April 2026)

I have a paid consulting relationship with Verasight (disclosed on my user page), so I am proposing this for editor review rather than editing the article directly.

Would editors consider whether a short, attributed public-opinion sentence belongs in a section discussing public perception of AI leadership or governance? The article is high-level, so I defer to editors on whether this is too narrow for this page or would fit better in a more specific article.

Proposed text: An April 2026 Verasight survey of U.S. adults found low public familiarity with several AI-related executives: 31% were not familiar with OpenAI's Sam Altman, 45% with Nvidia's Jensen Huang, and 50% with Anthropic's Dario Amodei.

Methodology: n=2,000 U.S. adults, fielded April 21-23, 2026, full-sample margin of error +/- 2.3 points. The item asked respondents their favorability toward named technology executives, with "Not familiar with this person" available as a response. I recognize this is a primary source and editors may want independent secondary coverage before including survey findings; if it is undue for this article, I understand.— Preceding unsigned comment added by SurveyDataNotes (talk • contribs) 17:14, 23 June 2026 (UTC)

 Not done - Thanks for cleanly disclosing your COI. But as you noticed, the fact that these are primary sources is an issue, and the popularity of AI CEOs is too tangential (WP:WEIGHT). Alenoach (talk) 18:56, 23 June 2026 (UTC)

Without context, which would have to come from a reliable, independent source, these factoids are trivia. Verasight doesn't appear to be a reliable source even if it weren't primary, and as you say, it is primary. You may continue to discuss this, if you absolutely insist, but do not re-open this template until consensus has changed. Grayfell (talk) 18:59, 23 June 2026 (UTC)

References

When and why was the citation format changed?

All the references were moved to "further reading". Not sure why. Do we need a discussion about the citation format? This article (mostly by default) uses a "hybrid" style to accomodate several different types of sources. We need shortened footnotes for sources used a lot and there are also citations-in-footnotes for sources used once. Shortened footnotes need a reference list. ---- CharlesTGillingham (talk) 18:30, 15 September 2026 (UTC)

Even at a glance, I see far, far too much editorializing and WP:OR/WP:SYNTH in these extended footnotes. Whatever formatting is used, the article needs to neutrally summarize reliable, independent sources about AI. It cannot use footnotes to justify publishing novel conclusions about related topics. The formatting issues are secondary to the content issues. Grayfell (talk) 22:38, 15 September 2026 (UTC)

I was just talking about the difference between References and Further Reading -- all the sources had been moved, which causes a bunch of errors to show up on my screen.

I assume you're talking about the "explanatory notes". Which ones, in particular, are OR or SYNTH? ---- CharlesTGillingham (talk) 10:44, 16 September 2026 (UTC)

Now I see. I agree that there was WP:SYNTH in the paragraph you removed --- my bad; Christensen was WP:SYNTH. I agree with all your other edits as well.

I disagree that the point is irrelevant. I may re-add some version of "previous waves of automation have tended to increase employment" with only R & N as a source. (They have a section that makes this point.)

I think you're wrong about the explanatory footnotes, if that's what you were talking about -- they all look reliable to me -- half of them are quotes from hyper-reliable sources. ---- CharlesTGillingham (talk) 11:01, 16 September 2026 (UTC)

Sorry if I was being dismissive. Ref formatting does matter, but my experience has been that discussions of it can quickly degrade into bikeshedding. I'm guilty of that, but regardless, it wouldn't be an excuse to ignore the problem.

Per H:EFN, explanatory notes should be used sparingly, and their usage here doesn't seem sparing. Beyond the accessibility issues, it just makes the article a little more tedious to read. There is also a more subtle issue. The article includes too many subtle vague and editorializing words, and the use of EFN superficially imply support while bypassing the underlying problem. WP:HOWEVER, "greatly", "extensively" "significant" (not in a statistical context), "increasingly" etc. Taken one-by-one, none of these are a problem in isolation, but overall, the article ends up using a lot of words to avoid making falsifiable claims. In my opinion, the many EFN make this worse, regardless of how reliable the sources would be if used in the body of the article.

Maybe others do not see it this way, which is fine, but of course it's worth thinking about what these sources are doing in the article first, before figuring out how to organize them. Grayfell (talk) 06:23, 17 September 2026 (UTC)

Feel free to fix the editorializing wherever it appears. Please be careful not to distort the meaning or compromise readability. Be sensitive to cases where the paragraph is reporting an opinion of experts, and the "editorializing" is coming from the source, not from Wikipedia. Don't just delete words. Find other ways to report the nuance in the source.

I'm not sure what you're seeing in EFN. There must be an example you have in mind. Which footnote(s) are you talking about?

In fact concrete examples would help a lot here. Just fix them and I'll review. ---- CharlesTGillingham (talk) 17:11, 18 September 2026 (UTC)

(By the way, there are sections (like Applications or Regulation) that have not been reviewed by me ... so if that's where the problem is, then ignore everything I've said. These sections don't have EFNs.) ---- CharlesTGillingham (talk) 17:27, 18 September 2026 (UTC)

The accessibility issue with this number of EFNS still remains, regardless of anything else.

I have been working to fix the article, but it's a massive task.

Artificial intelligence#Deep learning seems like a good example of what I'm talking about.

Deep learning has profoundly improved the performance of programs... - The cited source is from 2012 (). It's insufficient for such broad claims, and the language isn't neutral either way. This is using one example for claims about the entire subfield, which appears to me to be WP:OR.

The reason that deep learning performs so well in so many applications is not known as of 2021 - This phrasing is loaded.

The sudden success of deep learning in 2012–2015 did not occur because of some new discovery... - A source from 2012 cannot support sudden improvements in the three years after the source was published. None of the sources in the two EFNs that follow directly support this statement, and there are no other citations here. All of the EFN sources are historical. The one source from a contemporary work is the quote from The Alignment Problem about the 1990s, but this doesn't directly support this either. This is using old sources to make claims about recent developments. This is a form of SYNTH where sources have been tucked away in an EFN.

I am not disputing that there were improvements, but cited sources don't support this, and this is obfuscated by the use of EFNs.

Opinions from experts are still opinions, and need to be attributed. If an opinion is in an EFN, but the point in the body is not presented as an opinion, or worse, isn't directly supported by sources at all, that's a problem for the article. Grayfell (talk) 20:23, 18 September 2026 (UTC)

This paragraph is indeed a good example of problematic sourcing. I checked the sentence "The reason that deep learning performs so well in so many applications is not known as of 2021", and I don't find support for this in the page 750 of the book (either 3rd or 4th edition). It doesn't fully address your comment, but I made an edit removing the sentence and adding an inline tag. Alenoach (talk) 00:56, 19 September 2026 (UTC)

The reason that deep learning performs so well in so many applications is not known as of 2021 AIMA 2021, p. 750: "the true reasons for the success of deep learning have yet to be fully elucidated". (You must have missed it.) It's a paraphrase of a sentence in the most reliable source we have. I can't imagine how it could be better sourced. (BTW, this sentence did not use an EFN.) ---- CharlesTGillingham (talk) 18:28, 21 September 2026 (UTC)

You're right, I got confused by the page numbering (in my AIMA 2021 pdf version it's p. 801 but I assume that in print version or something it's p. 750). I still think the statement is too broad (maybe there is no consensus on the main reasons why it works so well, but there surely are some partial or contested explanations). But it is sourced, so I let you decide what to do with that, if you want to restore as it was you can. Alenoach (talk) 23:32, 21 September 2026 (UTC)

Deep learning has profoundly improved the performance of programs in many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing, image classification,} Agree that this had the wrong source somehow. Removed the source and added a "citation needed". I'm sure a solid source can be found. Stand by. (This also did not use an EFN) ---- CharlesTGillingham (talk) 18:53, 21 September 2026 (UTC)

The sudden success of deep learning in 2012–2015 did not occur because of some new discovery... Christian's book was published in 2020, after the decade that the deep learning revolution occurred, and has a detailed history of the period. (I would also argue that it is a maximally-reliable tertiary source). But, if it helps, I found a more recent source.

This long sentence does contain material that is not in the source, but this is not synthesis. It just explains in a bit more detail exactly what we're talking about. This could use sources, I suppose, so I added "citation needed" (even though I find it unlikely this would be challenged --- you guys need to be clear if you're challenging these points). Sources can be found: on Moore's Law, that GPUs were the innovation that allowed AlexNet to win the ImageNet competition in 2012, and the role of "big data" in the deep learning revolution, and the importance of curated data sets in the early 2010s.

The text-source integrity is a bit wonky and maybe that can be improved. I moved the Hinton quotes before the details, but this is still wonky. I also over-sourced by adding a second quote. Pick which quote you like and remove the other. (This quote is in an EFN, but here the EFN is functioning as a short cite with a verifying quote, which is a normal practice for points that have been challenged.) Finally, the footnote here should be bundled. I'll get to that. ---- CharlesTGillingham (talk) 20:28, 21 September 2026 (UTC)

On the explanatory-footnote issue, readability and source should not be competing priorities. As editors we can work to make sure that readability and verifiability work together (see Wikipedia:Verifiability). The main text ought to be understandable without requiring the reader to work through a large number of footnotes. Help:Explanatory notes says to use explanatory notes sparingly, which seems relevant here.

I think it's important to preserve useful context. If a footnote contains info that is necessary to understand or substantiate a claim in the body text, it's probably a sign that the sourcing/writing needs to be clearer. The article should explain a point clearly and cite it directly. Footnotes are useful for supplementary context that an inquisitive reader can explore further). Sergeant Curious (talk) 20:28, 22 September 2026 (UTC)

<< Okay, I think I fixed the text-source integrity (by removing all the parenthetical asides). Found sources for the final two points. Still need a source for the first sentence, shouldn't be hard to find. ---- CharlesTGillingham (talk) 22:16, 21 September 2026 (UTC)

I have further edited this section for brevity. We do not need two vague quotes from Hinton making basically the same point, we shouldn't editorializng to say that something is 'very important' without indicating why it is important, or at least who thinks it's important. We also shouldn't hide the history of the topic in a footnote. For convenience, here is a copy/paste of this version:

Deep learning has been used to improve the performance of programs in subfields of artificial intelligence including computer vision, speech recognition, natural language processing, image classification and others.

Deep neural networks and backpropagation had been in development since 1950.

Starting in 2012, the speed of deep learning was increased one hundred-fold by switching to GPUs, and there were enormous amounts of data available on the internet (called at the time "big data") as well as curated datasets used for benchmark testing, such as ImageNet. Usage of deep learning increased in 2012–2015 due to these improvements. In 2025, Geoffrey Hinton said that, until the 2010s, "We couldn't do anything very impressive because we didn’t have enough data and we didn't have enough computation."

References

Notes

The first paragraph is unchanged, so I have omitted it here. For the rest, I have attempted to cut out some redundancy and loaded wording, while still conveying the gist. Hopefully this also demonstrates what I mean when I say that EFNs have made parts of the article more confusing than it needs to be. Grayfell (talk) 07:33, 23 September 2026 (UTC)

Several things:

(1) The EFN you removed was just a citation with a quote. (i.e., it's just like a CS1 citation with the |quote= parameter set.) Is this what bothers you? This is a standard practice in Wikipedia, so I'm not sure why it bothers you. This is just a way of adding an extra layer of verification; a way of proving that the material is not "editorializing", but rather is from a reliable source. There are a dozen or so EFNs like this. Is this what you're talking about? If so, then I disagree strongly.

(2) This material now seems kind of incoherent to me. You've removed the topic points: no theoretical breakthrough, no theoretical explanation why it works so well, just more data+computation. If this isn't what we're the paragraph is talking about, then I don't know what it's trying to say. Every good paragraph requires a topic sentence, and you've removed it. What do you think it should be?

(3) Do you all object to reporting these points about deep learning? The leading AI textbook puts them in the second third paragraph of it's introduction to the subject. The Nobel laureate "Godfather" of deep learning makes these points in almost every interview. Why should we leave it out? To me, leaving it out feels like "editorializing" -- like we're trying to hide something here, something the most reliable sources think is important.

That's my thoughts. I won't edit war here. ---- CharlesTGillingham (talk) 02:54, 28 September 2026 (UTC)

I accept that my version may be kind of incoherent, but as I see it, it was already like that before I came along, just with more words. The previous version's explanation of these breakthroughs was not clear, nor was it neutral. If the godfather of AI calls them "breakthroughs", attribute that with enough context, but even so, it would be so much clearer if we used some of these many, many reliable sources to explain what this breakthrough means in falsifiable terms. If this cannot be put into falsifiable terms, that suggests deeper problems.

So the formatting is not the problem by itself, it's that the formatting made the sources and subtle editorializing language harder to evaluate. For example, exactly how "profoundly" did it improve programs in "important" subfields, and exactly how important are those subfields? I'm confident this can be explained clearly and neutrally without this style of editorializing, even if my version is also flawed. Grayfell (talk) 05:52, 5 October 2026 (UTC)

<< @Grayfell: I think I have a solution, and I’ll describe that below, but first allow me a rant.

“Falsifiability” is not a criteria for inclusion in Wikipedia. ‘’Verifiability’’ is the criteria. This applies to individual facts but also (in my view) to editorial choices about relevance and notability. We should report what the most reliable tertiary sources think is important, not what we think is important.

We’re discussing whether the article should report this fact: Deep learning outperforms other methods and there was no widely accepted theoretical explanation why this is true as of 2022.

This is clearly not “editorializing”, because it is drawn from maximally reliable sources without any OR or synthesis. It's a paraphrase. These facts are also considered highly relevant by the most reliable sources. So they pass any reasonable standard for “verifiability”, “relevance” and “notability”.

So I’m not sure I understand why you consider these facts “unfalsifiable”. More importantly, I’m not sure if that matters. The only question at issue is always “is this an accurate paraphrase/summary or not?” Do you argue that Russell & Norvig don’t say these things? Do you disagree that these are points that they make early in their introduction to deep learning? If so, then we can argue about it, and I know how to win that argument. (I also know how to admit I was wrong.)

But I don’t know how to win my argument with you, because I’m not 100% sure what you are talking about. “Falsifiability” is an epistemic quality that relates to the truth of the statement. We’re not in the business of “truth” here. If Russell and Norvig say something that’s not “falsifiable”, that’s not really our problem. We just report what they say.

But maybe I don’t understand exactly what you are trying to do. I can’t tell if you dispute the fact above or not. I can’t tell if you understand that it has a reliable source or not. I can’t tell what you don’t like about it. So forgive me if I’m not following what you’re trying to do. If I understood it, maybe we would be on the same page.

That’s the rant.

I hope it’s clear that I am a fanatic about verifiability, including verifiable relevance. (That might be the reason that this article is so heavily sourced.) I am also a fanatic about coherence and readability. Sections should have topics, paragraphs should have topics, one sentence should follow from the previous, only relevant sentences should appear and unnecessary sentences (or paragraphs or sections) should be cut for brevity or banished to a footnote. Wikipedia is not a loose collection of facts. (This is very hard to enforce here, because there are so many drive-by contributions, fixing it is hard, and I don’t have the time.)

Anyway I have a solution. I’m moving the “deep learning revolution” stuff down into the history section (including the detailed footnote on the origin of backpropagation). That section already had "reliable sources to explain what this breakthrough means in falsifiable terms."

I'm adding back in the fact above as stated by Russell and Norvig (in a quote-supported citation in a footnote for maximum verifiability). I'm using parenthetical referencing in a footnote, rather than {{efn}} to remove the extra level of indirection. <EDIT>

I may not finish this change today. ---- CharlesTGillingham (talk) 17:50, 8 October 2026 (UTC)

Nevermind, got it done. ---- CharlesTGillingham (talk) 20:05, 8 October 2026 (UTC)

Recent comments by AI executives

Recent Comments made on the future AI and its danger by the chat gpt ceo and others should be included Trivelaspin (talk) 09:08, 19 September 2026 (UTC)

The shitstorm of Sept. 2026 could be covered in the section "Regulation". This section has a lot of material that seems to be out-of-date, which should be cut down or tossed. ---- CharlesTGillingham (talk) 03:39, 22 September 2026 (UTC)

Superintelligence

After the UN speech by Trump, User:Cube1ber first immediately moved this page to Super intelligence (SI), and when this was reverted repeatedly added 4K of prose about this to the article. While this may warrant one or two sentences in this article at most, in the end how Trump wants to call AI is not important for an encyclopedic article about the large subject AI with its long history. Fram (talk) 16:08, 22 September 2026 (UTC)

Good reverts. Dedicating more than a sentence to this would be a waste. Renaming the article would be downright silly. ApLundell (talk) 17:41, 22 September 2026 (UTC)

It's not even worth a sentence until at least something official is in place. Large-scale changes would require the industry to adopt the name SI, however. RammyRamRamRam (talk) 18:19, 22 September 2026 (UTC)

I just wanted to thank those who caught the attempt to change the name of this Wikipedia article. You all acted rapidly and shut down the attempt quickly... You Editors deserve a medal for defending Wikipedia. Good Job! MagnummSerpentinee (talk) 20:59, 22 September 2026 (UTC)

The U.S. State Department ordered its international diplomats to use "Super Intelligence" ("SI") instead of AI. However, per Wikipedia's article naming policy, this article should not be renamed unless an alternative name becomes prevalent in reliable sources. GN22 (talk) 23:45, 24 September 2026 (UTC)

According to Trump, he and the President of China Xi Jinping agreed to refer to AI as "super intelligence" or "SI". However, the news media continues to refer to it as "AI". GN22 (talk) 19:39, 26 September 2026 (UTC)

I think that can all be safely ignored, especially there already is a concept of Super Intelligence. MrMarmite (talk) 23:53, 26 September 2026 (UTC)

This is the English Wikipedia, used worldwide. It is not a department of the U.S. government. It should use the world's terminology. ---- CharlesTGillingham (talk) 00:40, 29 September 2026 (UTC)

Great: now there's an executive order. May need a new page about the ensuing naming dispute. GN22 (talk) 03:32, 30 September 2026 (UTC)

Will OpenAI become OpenSI then? 😹 NectarLupine🪻~Talk 16:53, 4 October 2026 (UTC)

California's governor signed an executive order requiring all state executive agencies to reject Trump's rebranding and use "artificial intelligence". OpenAI's headquarters, as well as Anthropic's, Google's, and SpaceXAI's, are all in California… GN22 (talk) 17:21, 4 October 2026 (UTC)

Actually, Elon Musk is on board: SpaceXAI will become SpaceXSI. GN22 (talk) 17:51, 4 October 2026 (UTC)

Musk is Trump-aligned while the executives of the other big AI companies aren't. GN22 (talk) 17:57, 4 October 2026 (UTC)

Let us see what happens.... Once the hype dies off NectarLupine🪻~Talk 17:57, 4 October 2026 (UTC)

• This doesn't deserve this much discussion. The term AI predates the Internet, and even the ARPANet before that. These rebranding efforts will all fade away. O3000, Ret. (talk) 20:38, 4 October 2026 (UTC)

  It looks like several tech executives have adopted Trump's rebranding, including those of SpaceX, Salesforce, Amazon, and Microsoft. GN22 (talk) 16:26, 11 October 2026 (UTC)

  Amazon. AI ., Microsoft.AI , Salesforce. AI have no mention of anything other than AI. Ignoring the madness is the best way MrMarmite (talk) 16:39, 12 October 2026 (UTC)

COI request to better define "Firefly"

Hello, I'm a COI editor who works for Adobe. I was wondering if two instances of "Firefly" in this article could be changed to "Adobe Firefly." The first instance is in the Generative AI subsection and the second is in the big purple Artificial intelligence table at the bottom of the article. I'm asking for this change in order to clearly identify the product by using its full name (see: Adobe Firefly) and to distinguish it from the Firefly algorithm, which is also mentioned in this article. Thanks! Eric at Adobe (talk) 23:24, 13 October 2026 (UTC)

Done. Extraordinary Writ (talk) 05:00, 14 October 2026 (UTC)