AGI Zeitgeist

Panel evidence

Panel check C: the 2020s group and two skeptics

A working document. It is one of the documents the site is built from, published as it is written for anyone checking the work, and it uses the project's own shorthand. The plain account of the same material is on the home page.

Panel check C: the 2020s group and two skeptics

Checked 2026-09-30. Every quote below was copied from a page fetched in this session. All quotes come through the fetch tool's text extraction. Spot-check any quote against its page before it goes into data/statements.csv.

Limits of this check (read first).

Part 1: 2020s group candidates

Name Earliest statement (date, quote, link, primary/secondhand) Latest statement (date, quote, link, primary/secondhand) Clears core bar Notes
Dario Amodei 2017-07-21. "AGI is something that I think definitely will eventually happen." 80,000 Hours podcast #3, transcript (the date comes from the podwise episode listing). Primary (podcast transcript, Amodei speaking). 2026-01 (month only). "As I wrote in Machines of Loving Grace, powerful AI could be as little as 1–2 years away, although it could also be considerably further out." The Adolescence of Technology. Primary (essay). Yes The 2017 statement says AGI is possible and will happen. It gives no date. In the same answer he says he prefers "AGI" to "superintelligence" because no one knows whether a machine will "rocket past human level". His 2016 paper "Concrete Problems in AI Safety" (arXiv 1606.06565, 2016-06-21) is about safety only: the abstract covers "accidents in machine learning systems" and never mentions AGI, human-level AI or timing. It does not count. Other statements: 2024-10 "I think it could come as early as 2026, though there are also ways it could take much longer." (Machines of Loving Grace, primary). He moves to the core panel; his first date-type statement would be 2024.
Leopold Aschenbrenner 2023-03-29. "I think <1 year takeoff scenarios are pretty plausible (modulo coordination/regulation), meaning <1 year from human-level AGIs to crazy superhuman AGIs." Nobody's on the ball on AGI alignment. Primary (his blog). 2024-06 (month only). "AGI by 2027 is strikingly plausible." Situational Awareness. Primary (essay series). No The 2023 post is about how fast human-level AI would become far more capable, not when it arrives. It also says "If timelines are short and we don't get our act together, we're in a lot of trouble." I found nothing he has said publicly about timing after June 2024; since then press coverage has been about his investment fund. I did not search for his 2024 interview with the podcaster Dwarkesh Patel. 2020s group.
Daniel Kokotajlo 2020-08-23. "if transformative AI is achievable in the next five orders of magnitude of compute improvement (e.g. prosaic AGI?), it will likely be achieved in the next five years or so." His reply in the LessWrong "Forecasting Thread: AI Timelines". Primary (forum reply). 2026-08-16. His personal medians: automated coder "p50 Nov 2027"; artificial superintelligence "p50 Mar 2029". Q2.5 2026 Timelines Update (AI Futures Project blog). Primary (co-written blog post that states his own numbers). No His posts start in 2019. The 2019 posts I checked do not state when AI arrives. The fetched summary of his 2020 forecast says its median was about 2035; I did not see that number on the page myself. 2021-08-06 What 2026 looks like: in a 2022 comment there he says "When I wrote this story, my AI timelines median was something like 2029." "p50" means the median, his 50% date. 2020s group.
Ajeya Cotra 2020-09-18. "This predicts a median of 2052 for the year in which some actor would be willing and able to train a single transformative model." Draft report on AI timelines (LessWrong). Secondhand. The sentence is from Rohin Shah's newsletter summary attached to Cotra's post, not her own words. Her own account: 2022-08-02, "A median of ~2050 for transformative AI (30 years from posting, 28 years from now)" (Two-year update, primary). Recorded 2025-10-20, published 2026-02. Sources disagree on the publication day: the forum shows Feb 17, a search summary said Feb 24. "I think that probably in the early 2030s we are going to see what Ryan Greenblatt calls top-human-expert-dominating AI" 80,000 Hours podcast. Primary (transcript). No The 2022 update moved her median to "~2040". The report itself is a Google Drive document I did not open. 2020s group.
Jensen Huang 2023-11-29. "within the next five years, you're going to see, obviously, AIs that can achieve those tests." CNBC report, republished by NBC New York. Secondhand (news report quoting the New York Times DealBook conference). 2026-03-22 or 23 (the episode came out about Mar 22; the article is dated Mar 23). "I think we've achieved AGI" TechSpot, on the Lex Fridman podcast. Secondhand. The official transcript was cut off before that chapter when I fetched it. No No statement before 2018 turned up. One search for 2016–17 remarks found only 2024–26 coverage, so this is not proof that none exists. His AGI statements always depend on how AGI is defined ("tests" in 2023; a billion-dollar company in 2026). 2020s group.
Mustafa Suleyman 2016-12-05. "when we say it's 20 years out or decades way, we say that it so far out, we can't really measure it" TechCrunch (same quote in Android Authority). Secondhand (a reporter's direct quote from TechCrunch Disrupt London). 2025-11-14. "achieving human-level performance at most tasks, I wouldn't say all knowledge tasks. That feels quite likely in the next five years" Silicon Valley Girl podcast, transcript on singjupost. Third-party transcript of his own words, so close to primary. Yes TechCrunch also quotes him: "general learning systems are still decades away". He goes from "decades away" (2016) to "quite likely in the next five years" (2025), which makes him a good example of someone who changed his mind. Not used: the November 2025 Microsoft AI "humanist superintelligence" page (microsoft.ai/superintelligence) gives no timing. Late-2024 interview on The Verge's Decoder podcast ("five to seven years"): I saw it only in search snippets and never opened the page, so it is not used. He moves to the core panel.

Result for Part 1: Amodei and Suleyman both clear the core bar.

The other four stay in the 2020s group: Aschenbrenner, Kokotajlo, Cotra and Huang. No public statement on whether or when AGI arrives was found for any of them before 2020 (Kokotajlo's posts begin in 2019 but don't address it). For Huang, 2023 was the earliest.

Part 2: Skeptic balance

Name Earliest statement (date, quote, link, primary/secondhand) Latest statement (date, quote, link, primary/secondhand) Clears core bar Notes
Luciano Floridi (proposed) 2016-05-09. "True AI is not logically impossible, but it is utterly implausible." Aeon essay. Primary (essay). 2024-11-05. Article title: "Why the AI Hype is Another Tech Bubble", Philosophy & Technology (DOI 10.1007/s13347-024-00817-w; the date and title come from the Semantic Scholar record). Primary article, but only the title was checked. Springer's login redirect blocked the full text. Unclear (likely yes) The 2016 essay is a clear "possible in principle, but not going to happen" position, in the line of philosophers like Hubert Dreyfus. I could confirm the 2024 article exists, but I did not read a sentence of it about AGI. His recent arXiv papers on large language models (2025, 2506.10130) do not mention AGI in their abstracts. Next step: get a 2024 or later sentence from the full text of the hype article.
Toby Walsh (proposed) 2016-02-20. "I have argued that there are many reasons why we might never witness a technological signularity." ("signularity" is his typo.) "The Singularity May Never Be Near", arXiv 1602.06462, HTML version. Primary (paper). None found. His university homepage lists heavy media work through 2026, including a public lecture "The AI Race" (August 2026) and the book The Shortest History of AI (2025). It links to no article with a timing statement. Unclear The same paper also says "even without a technological singularity, we might still end up with machines that exhibit super-human levels of intelligence". So he doubts it is near, not that it is possible. His 2018 book 2062: The World that AI Made gives a year, which would place him clearly on the timing chart. Next step: one search for a 2024–26 interview or column (he writes often for The Conversation and the Guardian; the Guardian could not be fetched).
Oren Etzioni (backup) 2016-09-20. "In essence, according to 92.5 percent of the respondents, superintelligence is beyond the foreseeable horizon." MIT Technology Review. Primary (his own opinion piece). None found. His Wikipedia page cites nothing from 2023–26 on AGI timing. Medium returned an access error. Unclear The sentence summarizes his own survey of experts and agrees with it. The quote is about superintelligence.
Ernest Davis (backup) 2015-03-01. Paper "Ethical guidelines for a superintelligence", Artificial Intelligence (DOI 10.1016/j.artint.2014.12.003; the date comes from the Semantic Scholar record). Only the title and date were checked. The PDF could not be read and ScienceDirect and the ACM site returned access errors. 2025-01-05. "Hoping that reliable commonsense will simply 'emerge' from large databases and video has proven unrealistic, again and again." AI still lacks "common" sense, 70 years later (with Gary Marcus). Primary (co-written). Unclear (likely yes) The 2025 quote is about common sense, not AGI timing directly.
Melanie Mitchell (not proposed) Nothing found on or before 2017. 2025-09-15. "Lots of people have been saying this kind of stuff for decades" Magical Thinking on AI. Primary (her Substack). The fetch tool returned this as a fragment inside its own summary, so check the full sentence. No (as far as checked) She is a strong and prominent skeptic, but her public record found here starts in 2018–19. She fits the 2020s skeptic slot or a later-arrival note better than the core panel.

Why Floridi and Walsh

  1. The core list already has skeptics working in AI itself: Gary Marcus, Rodney Brooks, Yann LeCun and Andrew Ng. The gaps are a philosopher continuing Hubert Dreyfus's critique and an AI researcher outside the United States who gives an actual date. Floridi fills the first and Walsh the second.
  2. Floridi. His 2016 Aeon essay is the best-known recent restatement of the old philosophical argument: real AI is possible in principle but "utterly implausible". He keeps it up in 2024 with a hype-bubble paper in a leading journal on the philosophy of technology. This gives the "skeptical, and doubts it is even well-defined" category a voice that is not an engineer's.
  3. Walsh. His 2016 paper argues directly against the idea of a runaway "singularity". His 2018 book puts a year on human-level AI (2062, as the title suggests). He is one of the most-quoted AI scientists in the press in Australia and Britain and is still very active in 2026. A dated forecast plots directly on the "when did each person expect it" chart, which a skeptic who only says "not soon" does not.
  4. Why not the others.
    • Davis: his 2024-and-later statements are almost all co-written with Gary Marcus on Marcus's Substack, so including him would count Marcus twice.
    • Etzioni: prominent in 2016, but I could not find a statement from 2024 or later. He is the first backup if Walsh's 2024-and-later record does not hold up.
    • Mitchell: prominent, but no record on or before 2017 was found.
  5. Caveat. Neither proposed skeptic yet has a 2024-or-later statement checked against the full text: for Floridi only the article title was confirmed, and for Walsh nothing was found. Freeze them only after one more search each. If Walsh fails, use Etzioni. If Floridi's article turns out not to discuss AGI, use Davis and note that he overlaps with Marcus.

Queries run (all on 2026-09-30)

Web searches

  1. Dario Amodei 2017 interview 80,000 Hours AI timelines
  2. Mustafa Suleyman 2015 OR 2016 OR 2017 general artificial intelligence decades away interview
  3. Jensen Huang 2017 artificial general intelligence comment
  4. Melanie Mitchell 2017 OR 2016 OR 2015 general AI far away essay
  5. Dario Amodei 2026 AGI powerful AI timeline statement
  6. Mustafa Suleyman humanist superintelligence microsoft.ai post
  7. Leopold Aschenbrenner Situational Awareness "AGI by 2027 is strikingly plausible"
  8. Daniel Kokotajlo earliest post AI timelines LessWrong 2019 OR 2020
  9. Ajeya Cotra draft report on AI timelines biological anchors 2020 median 2050
  10. Ajeya Cotra 2025 AI timelines interview automate AI research
  11. 80000hours.org podcast Ajeya Cotra "use AI to make AI safe"
  12. Daniel Kokotajlo 2020 "my timelines" TAI median LessWrong
  13. "Forecasting Thread: AI Timelines" 2020 LessWrong Kokotajlo
  14. Kokotajlo "The date of AI Takeover is not the day the AI takes over"
  15. Kokotajlo 2025 OR 2026 AI 2027 updated timelines superhuman coder median
  16. AI Futures Project blog Kokotajlo timelines update 2026 "Daniel"
  17. greaterwrong "Forecasting Thread: AI Timelines" August 2020 Amandango
  18. Jensen Huang DealBook November 2023 AGI five years "fairly competitive"
  19. Jensen Huang Lex Fridman "we've achieved AGI" transcript
  20. Jensen Huang 2016 OR 2017 interview "general intelligence" OR "superintelligence" OR "singularity" Nvidia CEO
  21. Mustafa Suleyman AGI "next 10 years" OR "plausible but not probable" 2024 OR 2025 interview
  22. The Verge Decoder Mustafa Suleyman AGI transcript "five to seven years"
  23. Leopold Aschenbrenner 2025 OR 2026 statement AGI timeline interview investor letter

Searches that were refused because the budget had run out

Pages fetched directly (every link in the tables above, plus the ones below)

Rendered from docs/panel-evidence-2026-10/statements-cohort-and-skeptics.md at version a7bd17a.