Group e: panel checks under amendment A13 (2020s cohort tier)

Summary: Karnofsky and Cowen both have public AGI statements from before 2018, so both fail the cohort rule. Karnofsky's is a 2016 timing estimate and Cowen's is a 2005 doubt about Kurzweil's Singularity timeline. Cowen looks like a plausible core-panel candidate. Karnofsky is unlikely to qualify for the core panel unless someone finds national-press coverage of him. Leike and Mollick have no public AGI timeline statement before 2018 that I could find, so both pass the cohort rule. Their timing statements are thin, though: they mostly decline to give a date.

Holden Karnofsky

Tier checked: cohort_2020s C1 earliest: 2016-05-16 | timing | "I think there is a nontrivial likelihood of transformative AI within the next 20 years." | https://forum.nunosempere.com/posts/zmBnKnpECErWsCiYq/some-background-on-open-philanthropy-s-views-regarding (a mirror of the Open Philanthropy blog post "Some Background on Our Views Regarding Advanced Artificial Intelligence"; the original URL https://www.openphilanthropy.org/blog/some-background-our-views-regarding-advanced-artificial-intelligence now redirects to coefficientgiving.org, which returned 403) | primary text read on a mirror. A search-result summary gives the fuller wording "at least 10% with moderate robustness, and at least 1% with high robustness"; I did not see that on a fetched page. Not checked: his GiveWell post on the Singularity Institute (May 2012) may contain an earlier statement on whether AGI is possible. Later reference point: 2021-09-07 | "more than a 10% chance we'll see transformative AI within 15 years (by 2036); a ~50% chance we'll see it within 40 years (by 2060)" | https://www.cold-takes.com/where-ai-forecasting-stands-today/ | primary, fetched C2 latest: 2025-08-28 | "I kind of think when we get this human-level AGI, we have a good chance of being six to 12 months from something way, way, way more powerful." | https://podcast.clearerthinking.org/episode/277/holden-karnofsky-the-most-important-century | primary (podcast transcript), fetched. This is about how fast superhuman AI follows human-level AI, not when human-level AI arrives. His Carnegie paper of 2024-09-13 (https://carnegieendowment.org/research/2024/09/if-then-commitments-for-ai-risk-reduction) gives no date either ("the coming years could see fast and hard-to-predict changes in AI capabilities"). C3 pieces: none found. One search aimed at the New York Times, Wired and MIT Technology Review turned up only his own posts and community forums. Informational only for the cohort. Quota tags: frontier_lab: likely applies, but it is a judgment call. He sat on OpenAI's board of directors 2017-2021 (a governance role, not research) and has been a member of technical staff at Anthropic since January 2025, working on its responsible scaling policy and safety planning (https://en.wikipedia.org/wiki/Holden_Karnofsky, fetched 2026-09-30). Whether a board seat or a staff role counts as "senior" is for John to decide. | skeptic: no. His own 2016 piece describes moving from skeptic to advocate on AI risk, and in 2016 he put at least 10% on transformative AI within 20 years. | non_us_uk: no (GiveWell and Open Philanthropy, both United States). | philosopher: no. Verdict: fails the cohort rule. He made a dated AGI timing estimate in May 2016, before 2018. Core rule: he meets the early and late statement tests (2016 and 2025), but I found no national-outlet coverage from 2017 or earlier, so his core eligibility is unclear and probably fails on prominence. If he is moved to the core panel, he would count toward the frontier_lab quota. Queries: (1) Karnofsky 2016 "transformative AI" "within the next 20 years" "10%" Open Philanthropy; (2) Holden Karnofsky 2025 AGI timeline transformative AI by 2030 probability; (3) Karnofsky Anthropic 2025 interview "powerful AI" timelines years away; (4) "Karnofsky" artificial intelligence risk Open Philanthropy 2016 OR 2017 New York Times OR Wired OR "MIT Technology Review". 4 of 8 used.

Jan Leike

Tier checked: cohort_2020s C1 earliest: 2019-12-16 | possibility | "I don't think anyone really knows whether that's the case or not." (asked whether deep reinforcement learning will scale to AGI) | https://futureoflife.org/?p=39550 (Future of Life Institute, AI Alignment Podcast transcript) | primary, fetched. The earliest timing statement I found is in the OpenAI superalignment announcement of 2023-07-05, which Leike co-wrote: "While superintelligence seems far off now, we believe it could arrive this decade." I saw that only secondhand on https://www.longtermwiki.com/wiki/jan-leike, which wrongly credits it to an 80,000 Hours podcast; openai.com/index/introducing-superalignment returned 403. How hard I looked: two searches on his work before 2018 (his PhD thesis on general reinforcement learning, Australian National University 2016; his Cambridge talk of 2015-12-16; his 2017 papers on safety test environments and on learning from human preferences). These are technical work on AIXI, a theoretical model of an ideal general agent, and on reward learning. None of it says when AGI arrives. I did not read the full thesis text. C2 latest: 2024-05-17 | "Building smarter-than-human machines is an inherently dangerous endeavor." | https://decrypt.co/231294/openai-exec-quit-safety-took-backseat-shiny-products | secondhand (the article quotes his X thread), fetched. This treats smarter-than-human AI as coming, but it gives no date. I found no dated timing statement from 2024 or later. C3 pieces: none found. One search found only DeepMind blog posts (2017-06-12, 2017-11-28) and papers, not national-outlet pieces. Informational only. Quota tags: frontier_lab: applies. Research scientist at DeepMind 2016-2021; head of alignment and superalignment co-lead at OpenAI January 2021 to May 2024; at Anthropic since 2024 (https://www.longtermwiki.com/wiki/jan-leike, fetched 2026-09-30; roles also on https://en.wikipedia.org/wiki/Jan_Leike). | non_us_uk: does not apply as written, because he has no AGI statements from before 2017. He was outside the United States and United Kingdom during his PhD at the Australian National University (2014-2016, Australia; trained at Freiburg, Germany). | skeptic: no evidence either way from before 2017. | philosopher: no (computer science and machine learning). Verdict: passes the cohort rule. I found no public AGI timeline statement before 2018; the earliest is from 2019 and it is about possibility, not timing. Caveat: his timing statements overall are thin. The 2023 "this decade" line is co-written by him and OpenAI and was seen only secondhand, and the 2024 statement has no date in it. The full trace may find few rows that state a timeline. Queries: (1) Jan Leike AGI timeline prediction "Leike" superintelligence this decade 2024 OR 2025 interview; (2) Jan Leike 2016 OR 2017 talk human-level AI "Leike" general reinforcement learning AIXI thesis when AGI; (3) "Jan Leike" 2025 says AGI "years" alignment Anthropic automated alignment researcher timeline; (4) "Jan Leike" 2017 DeepMind "AI safety gridworlds" OR "human preferences" Wired OR "MIT Technology Review" OR "New Scientist". 4 of 8 used.

Ethan Mollick

Tier checked: cohort_2020s C1 earliest: 2023-01-21 | possibility (no date given) | "By definition, we literally cannot know what happens after the Singularity." | https://www.oneusefulthing.org/p/becoming-strange-in-the-long-singularity | primary, fetched. The post date comes from the fetch summary; it fits a newsletter started in late 2022, but it is worth a recheck. How hard I looked: one search on 2016-2017 writing about AI futures, human-level AI or the Singularity found nothing from before 2022. Several sources say ChatGPT turned him "from an AI-skeptic to an AI-believer" in late 2022. His work before then is on entrepreneurship and management research. Middle reference point: 2024-07-22 | "I can't tell you whether they are right, but I can tell you that they believe they are." (on AI lab insiders' AGI dates) | https://www.oneusefulthing.org/p/confronting-impossible-futures | primary, fetched C2 latest: 2025-04-20 | "Given the definitional problems, I really don't know, but I do think they can be credibly seen as a form of 'Jagged AGI.'" | https://www.oneusefulthing.org/p/on-jagged-agi-o3-gemini-25-and-everything | primary, fetched. This is about whether AGI is already here (o3 and Gemini 2.5). Also: 2025-01-10, "What concerns me most isn't whether the labs are right about this timeline" (https://www.oneusefulthing.org/p/prophecies-of-the-flood, fetched). C3 pieces: none found. Not searched for specifically, since prominence is informational for the cohort and nothing from before 2022 turned up at all. Quota tags: none apply. philosopher: no (PhD in management from MIT Sloan, 2010; https://en.wikipedia.org/wiki/Ethan_Mollick). frontier_lab: no. non_us_uk: no (Wharton, United States). skeptic: no statements from before 2017. Verdict: passes the cohort rule. I found no AGI statement before 2023. Caveat: he almost never commits to a date. His statements are about whether AGI is here and how unsure everyone is, so a full trace may produce mostly possibility and uncertainty codes rather than timelines. Queries: (1) Ethan Mollick 2016 OR 2017 artificial intelligence future human-level AI singularity Wharton; (2) Ethan Mollick "AGI" One Useful Thing 2023 "artificial general intelligence" when; (3) oneusefulthing.org Mollick 2023 post AGI superintelligence "don't know" timeline scenarios (limited to oneusefulthing.org). 3 of 8 used.

Tyler Cowen

Tier checked: cohort_2020s C1 earliest: 2005-09-21 | timing (doubts Kurzweil's Singularity timeline) | "I feel confident in my skepticism" | https://marginalrevolution.com/marginalrevolution/2005/09/the_singularity.html ("Is the Singularity Near?", Marginal Revolution) | primary, fetched. Ruled out after checking: a Marginal Revolution post of 2013-01-14 only quotes Bruce Sterling ("We're no closer to 'self-aware' machines than we were in the remote 1960s"), so it is not Cowen's own words. A search result credited a 2015 Edge essay on "naches" to Cowen, but Edge shows it is by Samuel Arbesman (https://www.edge.org/response-detail/26117). His posts of 2014-06-07 (Bostrom's book "self-recommending") and 2014-04-23 (his conversation with Nick Beckstead) say nothing on AGI timing. I found no passage from Average Is Over (2013) on human-level AI. C2 latest: 2025-04-16 | "I think it is AGI, seriously. Try asking it lots of questions, and then ask yourself: just how much smarter was I expecting AGI to be?" | https://marginalrevolution.com/marginalrevolution/2025/04/o3-and-agi-is-april-16th-agi-day.html | primary, fetched. Also seen in a search result: "Why I think AI take-off is relatively slow" (2025-02), https://marginalrevolution.com/marginalrevolution/2025/02/why-i-think-ai-take-off-is-relatively-slow.html (not fetched). C3 pieces (relevant if he is moved to the core panel; all three are reviews of Average Is Over, a book about machine intelligence reshaping work and income. Whether that counts as "about AI futures" is borderline and is John's call): 1. 2013-09-21 | The Economist | "The American Dream, RIP?" | https://www.economist.com/news/united-states/21586581-economist-asks-provocative-questions-about-future-social-mobility-american | listed on Wikipedia's Average Is Over page (fetched); article not fetched 2. 2013-10-02 | The Wall Street Journal | "Book Review: 'Average Is Over,' by Tyler Cowen" | https://www.wsj.com/articles/SB10001424052702303342104579097482945031804 | listed on Wikipedia (fetched); article not fetched 3. 2013-09-17 | The Daily Beast | "Welcome to Tyler Cowen's Future of Genius Machines" | http://www.thedailybeast.com/articles/2013/09/17/welcome-to-tyler-cowen-s-future-of-genius-machines.html | listed on Wikipedia (fetched); article not fetched. This is the closest to "about AI futures"; whether The Daily Beast counts as a national outlet is debatable. Others on the same Wikipedia list: National Review 2013-10-28 "A New World of Work"; Washington Examiner 2013-10-04; Slate 2013-09-26; NPR 2013-09-12; Washington Post 2013-09-16. Quota tags: skeptic: possible, but ambiguous. In 2005 he doubted Kurzweil's Singularity scenario, which put it about 40 years out (2005-09-21, https://marginalrevolution.com/marginalrevolution/2005/09/the_singularity.html). The doubt is general ("only one needs to fail"), not an explicit "not within decades", so treat it as a near miss unless more is found from before 2017. | non_us_uk: no (George Mason University, United States). | philosopher: no (economist by training and appointment, despite writing on moral philosophy). | frontier_lab: no. Verdict: fails the cohort rule. He doubted a near-term Singularity in writing in 2005. Core rule: plausibly meets it. He has the early statement (2005) and the late one (2025), plus several national-outlet reviews from 2013 that name him. Two things are uncertain: whether book reviews about machine intelligence and the economy count as "about AI futures", and the fact that his early statement is a short doubt rather than a worked position. He is a reasonable core candidate and might count toward the skeptic quota if more statements from before 2017 turn up. Queries: (1) Tyler Cowen Marginal Revolution singularity skeptical "human-level" artificial intelligence 2010..2017; (2) marginalrevolution.com Cowen Bostrom Superintelligence 2014 OR 2015 OR 2016 AI risk (limited to marginalrevolution.com); (3) Tyler Cowen Edge 2015 "What do you think about machines that think" answer; (4) edge.org response-detail Tyler Cowen 2015 machines that think naches (limited to edge.org); (5) "Average Is Over" Cowen singularity "strong AI" OR "human-level" OR "superintelligence" quote; (6) Cowen "I don't" singularity OR "artificial general intelligence" marginalrevolution 2011 OR 2012 OR 2015 OR 2016 OR 2017 (limited to marginalrevolution.com); (7) "Tyler Cowen" "Average Is Over" 2013 machines intelligent future review New York Times OR Economist OR Wired OR "New Yorker". 7 of 8 used.

Rendered from docs/panel-evidence-2026-10/a13-cohort-new.md at version a7bd17a.