Search for more 2020s cohort members, 2026-10-06

Asked for by me on 2026-10-06 ("Search first, then freeze"): one to three more cohort members before the freeze, found at no cost by the same method, with the rules exactly as written in docs/methodology-v2.md section 3:

C1: no public AGI timeline statement found before 2018.
C2: one dated 2024 or later.

Candidates were drawn from the wide list (the MIRI and AI Impacts predictions dataset and the full TIME100 AI lists read the same day) and from other prominent voices of the 2020s, preferring people who would broaden the cohort: skeptics, people based outside the United States and United Kingdom, and academics not at an AI company. Co-written statements were treated as in A14 ruling 3 (they count when the person is a named author). Quotes were checked in the raw page text unless marked otherwise. Checked by claude-agent.

Before this search the cohort was 9, all based in the United States or United Kingdom, and none a skeptic.

Liang Wenfeng (founder and chief executive of DeepSeek, China)

Verdict: PASS (thin). Thin because the statement is read in an English translation.

C2: 2024-07. Interview with Waves (暗涌), a sub-brand of the Chinese technology outlet 36Kr, published July 2024; English translation by ChinaTalk, 2024-11-27: https://www.chinatalk.media/p/deepseek-ceo-interview-with-chinas (fetched; quote checked in the raw page text). Secondhand (translation of his spoken interview).

Waves: How long until AGI is realized? [...] Liang Wenfeng: It could be two, five, or ten years–in any case, it will happen in our lifetimes.

A dated range, in his own words. The same interview: "Our ultimate goal is still to achieve AGI."
C1: none found. He co-founded the quantitative fund High-Flyer (幻方) in 2015; the Chinese-language results describe the fund moving its trading to deep-learning models in 2016, with no statement about AGI or its timing. DeepSeek was founded in 2023. His first public AGI remarks found are from 2023.
Tags: non_us_uk (Hangzhou, China). frontier_lab: yes, DeepSeek from 2023 (its R1 model of January 2025 was close to the best of its time; arguable, so marked for me). Not a skeptic (expects AGI within his lifetime). Not a philosopher.
Prominence (information only): profiled widely after January 2025 (a Reuters story of 2025-01-28 seen in search results).

Queries: Liang Wenfeng DeepSeek interview AGI "2, 5, or 10 years" OR "within our lifetime" 2024; 梁文锋幻方 2016 OR 2017 通用人工智能 OR AGI 采访.

Arthur Mensch (co-founder and chief executive of Mistral AI, France)

Verdict: PASS (thin). Thin because the statement is about whether AGI arrives, not when, and is secondhand.

C2: 2024-04. Told the New York Times in April 2024 (the Times is not reachable from here); read in Fortune, 2024-04-12, by Ryan Hogg: https://fortune.com/europe/2024/04/12/millennial-boss-french-openai-challenger-tech-ceos-obsession-ai-outsmarting-humans-very-religious-fascination-agi/ (fetched; quote checked in the raw page text). Secondhand.

"The whole AGI rhetoric is about creating God," Mensch told the New York Times. "I don't believe in God. I'm a strong atheist. So I don't believe in AGI."

This is a rejection of AGI arriving at all. It is counted as a timeline statement in the sense the cohort table already uses (Leike's 2024 pass rests on a statement with no date; Mollick's on "Jagged AGI"). If I reads "timeline statement" as requiring a date or horizon, Mensch becomes a near miss.
C1: none found. He was a doctoral student at Inria and NeuroSpin, Paris, 2015 to 2018, working on brain-imaging analysis; then a postdoctoral researcher at the École normale supérieure; Google DeepMind Paris from 2020 (Wikipedia, seen in search results). No AGI statement before 2018 turned up.
Tags: non_us_uk (Paris). frontier_lab: arguable, marked for me: research scientist at DeepMind 2020 to 2023 (a research role, not clearly senior) and head of Mistral from 2023 (a strong European lab, not clearly at the frontier). Skeptic in outlook (2024), but the skeptic tag as defined rests on statements from 2017 or earlier, so it is not applied. Not a philosopher.

Queries: Arthur Mensch Mistral AGI "creating God" 2024 interview; Arthur Mensch before Mistral 2016 OR 2017 artificial general intelligence statement PhD Inria.

Arvind Narayanan (professor of computer science, Princeton University, United States)

Verdict: PASS (thin). Thin because the statement is co-written and gives no arrival date for AGI.

C2: 2025-04-15. "AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence", by Arvind Narayanan and Sayash Kapoor, Knight First Amendment Institute, 2025-04-15: https://knightcolumbia.org/content/ai-as-normal-technology (fetched; quotes checked in the raw page text). Primary, co-written (he is a named author).

"we think that transformative economic and societal impacts will be slow (on the timescale of decades)"

"This is related to one definition of artificial general intelligence (AGI)—a unified system that is capable of performing all economically valuable tasks. According to the normal technology view, such sudden economic impacts are implausible. [...] This means that the goalpost of AGI will continually move further away as increasing automation redefines which tasks are economically valuable."

The essay calls "superintelligent" AI a notion "which we view as incoherent". A statement about whether and how fast AGI arrives, doubting it, with a timescale of decades for its effects.
C1: none found. The search for 2015 to 2017 statements on AGI, human-level AI or superintelligence returned only his work from 2024 on. No pre-2018 AGI statement turned up.
Tags: none of the four as defined. Skeptic in outlook (2025), but the tag rests on statements from 2017 or earlier. United States. Computer scientist, not a philosopher. No AI company role. He broadens the cohort as its first academic skeptic not tied to an AI company.
Prominence (information only): TIME100 AI 2023 (with Kapoor); New York Times Hard Fork podcast, 2025 (seen in search results).

Queries: Arvind Narayanan AGI "AI as normal technology" 2025 superintelligence timeline quote; Arvind Narayanan 2015 OR 2016 OR 2017 "artificial general intelligence" OR "human-level AI" OR superintelligence blog.

Near misses and leads not taken

person	why not added
Sayash Kapoor (Princeton, co-author with Narayanan)	Meets the rule on the same co-written essay, with no pre-2018 statement (he was a student). Not proposed: his only qualifying statement is the one Narayanan's pass rests on, so he adds no independent voice, and adding him would take the cohort to 13, above the range of 10 to 12. I may prefer him to one of the three
Abeba Birhane (Trinity College Dublin, Ireland; skeptic)	Would broaden the cohort most (outside the US and UK, skeptic, academic), but the search found only criticism of AI hype in general ("inflated optimism around AI"), no 2024-or-later statement on whether or when AGI arrives. Fails C2 as searched
Katja Grace (AI Impacts; TIME100 AI 2024)	Lead author of the 2017 expert survey "When Will AI Exceed Human Performance?" (arXiv 1705.08807, May 2017), which reports surveyed researchers' forecast dates for human-level machine intelligence, and ran AI Impacts' timeline work from 2015. Probably fails C1; whether a survey report counts as her own timeline statement is arguable and was not pursued, since the three above already fill the range
Yi Zeng, Andrew Yao, Xue Lan (China; TIME100 AI lists)	Not checked within this search; leads for a later amendment if the cohort is reopened
Emily M. Bender (University of Washington; TIME100 AI 2023)	Not checked within this search

Effect

Cohort 9 to 12, the top of its range. The three added bring the cohort's first members based outside the United States and United Kingdom (Liang, Mensch), its first skeptics in outlook (Mensch, Narayanan), and its first academic not tied to an AI company (Narayanan). All three are thin, for the reasons given.

Resolved at the freeze (2026-10-06, amendment A18)

I froze the panel the same day ("Freeze and merge") and chose three readings, recommended by Claude: Liang is tagged frontier_lab, Mensch is not; Mensch's April 2024 "I don't believe in AGI" counts as his 2024-or-later statement; Narayanan is taken, not Kapoor, and Kapoor stays on the wide list.

Rendered from docs/panel-evidence-2026-10/cohort-search-2026-10-06.md at version a7bd17a.