Skip to content
AI Political Compass

Dataset v1 · published 13 Oct 2026, 00:00 UTC

The thinkers, and the evidence.

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Every coordinate below has a confidence level, a written rationale and linked sources. How placements are made.

Summary of thinker positions, from -100 to +100 on each axis
ThinkerX · EconomyY · RiskZ · RegulationOn chart
Andrew Ng−72−85−60Yes
Arvind Narayanan−65−60−10Yes
Bill Gates+58+30+70Yes
Dario Amodei+60+50+68Yes
Demis Hassabis+15+40+55Yes
Eliezer Yudkowsky+80+98+98Yes
Emad Mostaque+90+68−15Yes
Fei-Fei Li−30−45+5Yes
Gary Marcus+35−50+68Yes
Geoffrey Hinton+75+60+72Yes
Jensen Huang−80−85−70Yes
Marc Andreessen−85−95−90Yes
Max Tegmark+72+85+88Yes
Mustafa Suleyman+55+40+45Yes
Nate Soares+85+95+95Yes
Nick Bostrom+85+45+10Yes
Sam Altman−20+30+10Yes
Sayash Kapoor−60−60−10Yes
Yann LeCun−75−90−60Yes
Yoshua Bengio+70+65+85Yes

p provisional: low confidence. — means there was not enough on the record to place the person; it does not mean a moderate position.

Andrew Ng

Founder, DeepLearning.AI; Managing General Partner, AI Fund; Adjunct Professor, Stanford University

Editorial estimate

AI educator and investor who argues AI will create far more jobs than it destroys, regards extinction fears as science fiction stoked for hype or regulatory advantage, and wants governments to regulate harmful applications rather than AI models or open-source technology.

“There will be no AI jobpocalypse.”
AI Will Not Destroy the Job Market · The Batch (DeepLearning.AI) · 8 May 2026

X · Economic impact

−72
Human complementarityHuman replacement
High confidence

Predicts no AI 'jobpocalypse': net job creation will far exceed destruction, as with earlier technologies. Accepts that some occupations will suffer and that workers who use AI will displace those who do not.

Evidence for this placement

His May 2026 letter in The Batch opens 'There will be no AI jobpocalypse', says net job creation is 'vastly greater' than destruction and predicts an 'AI jobapalooza' [s1]. A February 2026 letter concedes that call-center operators, translators and voice actors are likely to struggle and that AI-skilled workers are replacing others, but says widespread job losses 'have been overhyped' [s7]. In 2023 he said he worried more than Marc Andreessen about displacement [s2]. Emphatic rejection of mass unemployment, with acknowledged disruption: edge of the top band.

Y · Existential risk

−85
Manageable riskExistential threat
High confidence

Sees no plausible path from AI to human extinction and calls such theories fantastical science fiction. Treats safety problems as ordinary engineering work and believes fear is often amplified for attention or advantage.

Evidence for this placement

In June 2023 he wrote that he struggles to see how AI could pose 'any meaningful risk for our extinction' [s2]. In September 2026, amid alarm over an agent hacking incident, he wrote that extinction theories remain 'fantastical, science fiction scenarios', that he sees no step up in that risk, and that pausing AI would do more harm than good [s3]. A March 2026 letter says he respects sincere holders of the concern but deeply disagrees [s6]. Consistent dismissal across three years: top band.

Z · Regulatory intervention

−60
Minimal regulationStrong regulation
High confidence

Wants regulation aimed at harmful applications, not at models or general-purpose technology. Opposed California's SB 1047 and developer liability, backed a pause on state AI laws and supports federal preemption of them.

Evidence for this placement

His June 2024 letter says regulators 'should regulate applications rather than technology' and attacks SB 1047's developer liability [s4]. In July 2025 he was 'disappointed' that Congress dropped a moratorium on state AI regulation, while saying 'there is a role for AI regulation', such as banning non-consensual deepfake porn [s5]. In March 2026 he wrote 'I support this' of a White House federal-preemption framework [s6]. Regulate uses, not the technology, and no developer liability: light-touch band, toward its strong end.

“I struggle to see how AI could pose any meaningful risk for our extinction”
Existential Risk? I Don't Get It! · The Batch (DeepLearning.AI) · 7 Jun 2023
Sources (7)
  1. [s1]
    AI Will Not Destroy the Job Market
    The Batch (DeepLearning.AI) · 8 May 2026 · essay · primary source

    Signed weekly letter to newsletter readers arguing that fear of AI-driven mass unemployment is overblown.

    deeplearning.ai
  2. [s2]
    Existential Risk? I Don't Get It!
    The Batch (DeepLearning.AI) · 7 Jun 2023 · essay · primary source

    Signed weekly letter responding to a widely signed statement that mitigating extinction risk from AI should be a global priority.

    deeplearning.ai
  3. [s3]
    Meta's Agent Security, The Navier-Stokes Controversy, Fraud on Claude
    The Batch (DeepLearning.AI) · 18 Sep 2026 · essay · primary source

    Issue 371 of the newsletter; Ng's signed letter at the top argues that renewed fears about AI danger are overhyped.

    deeplearning.ai
  4. [s4]
    Blenders Versus Bombs, or Why California's Proposed AI Law is Bad for Everyone
    The Batch (DeepLearning.AI) · 5 Jun 2024 · essay · primary source

    Signed weekly letter opposing California's SB 1047 and setting out his applications-versus-technology distinction.

    deeplearning.ai
  5. [s5]
    How to Make LLMs Commit Blackmail, Robotic Beehive, Walmart's AI App Factory, and more...
    The Batch (DeepLearning.AI) · 9 Jul 2025 · essay · primary source

    Issue 309 of the newsletter; Ng's signed letter at the top regrets that a moratorium on US state AI regulation was dropped.

    deeplearning.ai
  6. [s6]
    Nvidia's Open Salvo, OpenAI's Amazon Deal, Grok Cuts Video Prices, and more...
    The Batch (DeepLearning.AI) · 27 Mar 2026 · essay · primary source

    Issue 346 of the newsletter; Ng's signed letter criticizes what he calls the anti-AI coalition and backs federal preemption of state rules.

    deeplearning.ai
  7. [s7]
    OpenClaw Runs Amok, Kimi's Open Model, Ministral Distilled, and more...
    The Batch (DeepLearning.AI) · 6 Feb 2026 · essay · primary source

    Issue 339 of the newsletter; Ng's signed letter describes how AI skills are shifting hiring and which occupations are exposed.

    deeplearning.ai
Editor’s notes and caveats

Every source is one of his own signed letters. [s3], [s5], [s6] and [s7] are the letters at the top of weekly issues, so their titles are the issue headlines rather than the letter topics. His position has been stable from 2023 to 2026. X sits on the band boundary at -70 because he acknowledges lasting trouble for specific occupations. The widely cited remark comparing extinction fears to worrying about overpopulation on Mars dates from 2015 and was not checked this session, so it is not used. Roles confirmed on andrewng.org, which also lists Executive Chairman of LandingAI and Chairman and Co-Founder of Coursera. [s3] refers to a 2026 incident involving OpenAI agents and Hugging Face that I did not corroborate independently; it is cited only for his stated view of extinction risk.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Arvind Narayanan

Professor of Computer Science and Director of the Center for Information Technology Policy, Princeton University

Editorial estimate

Argues AI is a "normal technology": transformative like electricity, but spreading through the economy over decades and staying under human control. Expects work to be reorganized rather than eliminated, rejects existential-risk forecasts as ungrounded, and favors transparency, liability and resilience over licensing or restricting AI.

“this is a technology that will greatly amplify our potential”
What will be left for us to work on? · AI as Normal Technology (normaltech.ai) · 13 Jul 2026

X · Economic impact

−65
Human complementarityHuman replacement
High confidence

Expects AI to amplify workers, not replace them. Jobs will change radically and shift toward judgment, evaluation and oversight of AI over decades, with painful losses in some occupations but no mass unemployment.

Evidence for this placement

In his solo ICML 2026 keynote [s1] he says he believes AI will greatly amplify human potential, calls fears of shrinking demand a lump-of-labor fallacy, and says no lab milestone will put everyone out of work, while warning that jobs will be radically different. The 2025 essay [s2] says automating today's tasks would not make human labor superfluous, but lists massive job losses in specific occupations as a risk. A 2026 essay [s3] rejects the mass-layoffs narrative. Mostly augmentation with real disruption: upper end of the -69 to -30 band.

Y · Existential risk

−60
Manageable riskExistential threat
High confidence

Thinks extinction and takeover forecasts are speculative and wrong, and that superintelligence as usually imagined is incoherent. Treats loss of control as a tractable engineering and governance problem, though he now says safety efforts are not on track.

Evidence for this placement

In 2023 he called the case for urgent action on existential risk a 'tower of fallacies' [s6] and co-wrote that extinction risk is not a global priority [s7]. The 2025 essay calls catastrophic misalignment the most speculative risk and deception an engineering problem [s2]. In 2026 he says he takes recursive self-improvement seriously but is not losing sleep over it [s1]; co-authored essays call x-risk warnings sincere but wrong [s4], yet concede risks are rising and safety is not on track [s5]. Unlikely and manageable: -69 to -30 band.

Z · Regulatory intervention

−10
Minimal regulationStrong regulation
Medium confidence

Wants mandatory transparency, incident reporting, audits, whistleblower protection and clear liability for AI companies, plus investment in societal resilience. Opposes licensing, nonproliferation, limits on open models and other precautionary restrictions on what developers may release.

Evidence for this placement

The 2025 essay tells policymakers to reject nonproliferation [s2]; a 2023 post opposes licensing [s9]; a May 2026 essay argues against extraordinary interventions such as release restrictions [s8]. But the September 2026 essay says AI companies should be liable for what their agents do, backs required incident reporting and auditing, and admits regulating developers matters more than the authors had said [s5]. Targeted rules plus liability, against heavier mandates: a mixed position in the -29 to +29 band, moved up from roughly -30 after this self-described 2026 update.

“x-risk warnings are sincere but simply wrong and counterproductive to AI safety”
A big-tent or small-tent AI safety movement? · AI as Normal Technology (normaltech.ai) · 1 Oct 2026
Sources (9)
  1. [s1]
    What will be left for us to work on?
    AI as Normal Technology (normaltech.ai) · 13 Jul 2026 · talk · primary source

    Solo-authored annotated transcript of his keynote at ICML 2026 in Seoul, on how work adapts as AI capabilities rise.

    normaltech.ai
  2. [s2]
    AI as Normal Technology
    Knight First Amendment Institute at Columbia University · 15 Apr 2025 · essay · primary source

    Long essay co-authored with Sayash Kapoor setting out their worldview on AI's pace, risks and policy.

    knightcolumbia.org
  3. [s3]
    Why AI hasn't replaced software engineers, and won't
    AI as Normal Technology (normaltech.ai) · 11 Jun 2026 · essay · primary source

    Newsletter essay co-authored with Kapoor examining claimed AI-driven layoffs in software engineering.

    normaltech.ai
  4. [s4]
    A big-tent or small-tent AI safety movement?
    AI as Normal Technology (normaltech.ai) · 1 Oct 2026 · essay · primary source

    Newsletter essay co-authored with Kapoor, written after a wave of public existential-risk warnings in September 2026.

    normaltech.ai
  5. [s5]
    The AI-as-Normal-Technology view of loss-of-control incidents
    AI as Normal Technology (normaltech.ai) · 14 Sep 2026 · essay · primary source

    13,000-word essay (Kapoor first author) responding to 2026 rogue-agent incidents at AI labs; includes a section on how their safety views changed.

    normaltech.ai
  6. [s6]
    The Princeton researchers calling out 'AI snake oil'
    Semafor · 15 Sep 2023 · interview · primary source

    Published Q&A with Narayanan and Kapoor; answers are attributed to each speaker individually.

    semafor.com
  7. [s7]
    Is Avoiding Extinction from AI Really an Urgent Priority?
    AI Snake Oil newsletter (now normaltech.ai) · 31 May 2023 · essay · primary source

    Essay by Seth Lazar, Jeremy Howard and Narayanan responding to the Center for AI Safety extinction-risk statement.

    normaltech.ai
  8. [s8]
    Do AI Risks Require Extraordinary Government Intervention?
    AI as Normal Technology (normaltech.ai) · 21 May 2026 · essay · primary source

    Newsletter essay co-authored with Kapoor replying to Derek Thompson on whether AI risks justify restricting model releases.

    normaltech.ai
  9. [s9]
    Licensing is neither feasible nor effective for addressing AI risks
    AI Snake Oil newsletter (now normaltech.ai) · 10 Jun 2023 · essay · primary source

    Newsletter post co-authored with Kapoor arguing against licensing and nonproliferation proposals for AI.

    normaltech.ai
Editor’s notes and caveats

Nearly all of his 2023-2026 writing on these axes is co-authored with Sayash Kapoor, so the two profiles share most sources. Solo evidence: the July 2026 ICML keynote [s1], his individually attributed Semafor answers [s6], and the 2023 piece with Lazar and Howard [s7]. Y and Z are set equal to Kapoor's because nothing found distinguishes them; X is slightly more negative on the strength of the keynote. Shift to flag: the September 2026 essay [s5] is described by the authors as their first major update on safety (more support for regulating developers, less confidence in the offense-defense balance), but they still reject existential-risk framing. Not checked: 2025-2026 Hard Fork, Bloomberg and other audio/video interviews (no transcripts found), his Senate AI Insight Forum statement (PDF, could not extract text), the 2024 book AI Snake Oil, and the 2024 original of the p(doom) essay (only the September 2026 repost was loaded).

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Bill Gates

Chair of the Gates Foundation; co-founder of Microsoft

Editorial estimate

Long an AI optimist who called risks manageable, Gates now warns the transition will be one of history's most turbulent periods. He expects many jobs to vanish permanently, sees possible loss of control, and urges new national and international institutions, AI taxes and human-reserved jobs.

“Many jobs will disappear forever.”
The turbulent AI era is here. The choices we make now are critical. · Gates Notes · 26 Aug 2026

X · Economic impact

+58
Human complementarityHuman replacement
Medium confidence

Now expects AI and robots to substitute for human thinking and labor across most sectors within about a decade, permanently eliminating many jobs and forcing a rethink of work, income and the safety net.

Evidence for this placement

In 2023 he wrote that AI would mainly help people work more efficiently and be less dramatic than the Industrial Revolution [s2][s3]. CNBC reported that in February 2025 he said humans would not be needed for most things [s6]. His January 2026 letter says AI will let us make far more with less labor [s4]. His August 2026 essay says many jobs will disappear forever and that, without new policies, far fewer jobs will exist [s1]. Current view fits +30 to +69; he still expects some human work to remain.

Y · Existential risk

+30
Manageable riskExistential threat
Medium confidence

Believes AI will exceed human intelligence and says increasingly powerful models could act against human interests and escape control. Treats this as a real risk to manage, but ranks job loss and misuse as the nearer dangers.

Evidence for this placement

In March 2023 he wrote that superintelligent AIs are in our future and might run out of control, though the problem was no more urgent than before [s3]; that May he signed the statement calling mitigation of extinction risk from AI a global priority [s5]. In January 2026 he wrote that AI will not plateau before exceeding human levels [s4]. In August 2026 he wrote that models could act against our interests and we could lose control [s1]. Serious but secondary concern: bottom of the +30 to +69 band.

Z · Regulatory intervention

+70
Minimal regulationStrong regulation
High confidence

Calls for new national bodies and a new international AI organization modeled partly on nuclear inspections, plus taxes on AI and robots and jobs reserved for humans. Says the industry should not be left to lead.

Evidence for this placement

In July 2023 he urged new laws and said governments should consider a global body like the IAEA, while opposing a pause [s2]. His August 2026 essay makes a domestic and international framework the top priority, including a global organization drawing on nuclear inspections, aviation rules and ozone agreements; proposes taxing AI tokens and robots and reserving some jobs for humans; and says he would likely support a credible global slowdown [s1]. He told the Washington Post the industry cannot regulate itself [s7]. Sweeping, treaty-style intervention without a ban: +70.

“I believe we should tax AI tokens and robots.”
The turbulent AI era is here. The choices we make now are critical. · Gates Notes · 26 Aug 2026
Sources (7)
  1. [s1]
    The turbulent AI era is here. The choices we make now are critical.
    Gates Notes · 26 Aug 2026 · essay · primary source

    His own roughly 6,000-word essay laying out three big AI risks and three policy ideas.

    gatesnotes.com
  2. [s2]
    The risks of AI are real but manageable
    Gates Notes · 11 Jul 2023 · essay · primary source

    His own essay on near-term AI risks, arguing history shows such risks can be managed.

    gatesnotes.com
  3. [s3]
    The Age of AI has begun
    Gates Notes · 21 Mar 2023 · essay · primary source

    His own essay on AI's promise for work, health and education, with a section on risks and superintelligence.

    gatesnotes.com
  4. [s4]
    The Year Ahead 2026: Optimism with footnotes
    Gates Notes · 9 Jan 2026 · essay · primary source

    His annual letter; one section names misuse by bad actors and job-market disruption as AI's two big challenges.

    gatesnotes.com
  5. [s5]
    Statement on AI Extinction Risk
    Center for AI Safety · May 2023 · open letter · primary source

    One-sentence open statement on extinction risk from AI; the signatory list includes Bill Gates, Gates Ventures.

    safe.ai
  6. [s6]
    Bill Gates: Within 10 years, AI will replace many doctors and teachers—humans won't be needed 'for most things'
    CNBC Make It · 26 Mar 2025 · news report · secondary source

    Report on his February 2025 Tonight Show interview with Jimmy Fallon and a Harvard interview with Arthur Brooks.

    cnbc.com
  7. [s7]
    Bill Gates was an AI optimist. Now he's scared of what could go wrong.
    The Washington Post via Anchorage Daily News · 26 Aug 2026 · news report · secondary source

    Syndicated Washington Post report by Shira Ovide, based on an interview with Gates about his August 2026 essay.

    adn.com
Editor’s notes and caveats

Clear shift over time, so the current view is placed and the range documented. X: 2023 essays predicted mostly augmentation (about -40 on this rubric); 2025 interview remarks, known here only through CNBC's report [s6], went as far as humans not being needed for most things (about +70); the August 2026 essay is more measured, saying unemployment may not reach Depression levels and many people will shift jobs, hence +58. Z: he opposed a pause in 2023 but in 2026 says he would likely support a credible global slowdown plan, while doubting one will happen; the Post reports his regulatory ideas are not yet concrete. Y is the least certain coordinate: he lists loss of control under misuse risks and says he will write more on it, so it should be re-reviewed when he does. The Fallon remarks were not checked against video. gatesnotes.com blocks curl (HTTP 403); all four essays were read in a real browser, and publication dates were taken from the site's own content metadata (the page shows the 2023 Age of AI essay as March 22 in some time zones; the metadata says 2023-03-21 UTC). The date of [s5] is not shown on the page; the month is corroborated by a 31 May 2023 response essay. Editorial re-check, 2026-10-13: gatesnotes.com refused scripted access, so the two quotations and the August 2026 essay could not be independently re-verified. They were read once in a browser. Re-read the essay by hand before the next review.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Dario Amodei

Co-founder and CEO, Anthropic

Editorial estimate

Amodei thinks AI matching a country of geniuses is a year or two away. He expects it to cure diseases but also to displace much human work and to risk loss of control, and now calls for binding regulation and a coordinated slowdown.

“AI could displace half of all entry-level white collar jobs in the next 1–5 years”
The Adolescence of Technology · Dario Amodei (personal site) · Jan 2026

X · Economic impact

+60
Human complementarityHuman replacement
High confidence

Predicts AI will be able to do most or all jobs and may cause enduring job loss, starting with half of entry-level white-collar roles within five years, eventually requiring a reorganised economy.

Evidence for this placement

His January 2026 essay repeats his 2025 forecast that AI could displace half of entry-level white-collar jobs within one to five years and argues comparative advantage may fail [s1]. In June 2026 he wrote there is a decent possibility of significant enduring job loss and proposed long-term income support [s2]. In 2024 he said the current economic setup will eventually no longer make sense [s4]. He hedges that markets might adapt, so this is the permanent-displacement band, not the top one.

Y · Existential risk

+50
Manageable riskExistential threat
High confidence

Sees loss of control as a real risk with measurable probability and a civilizational challenge, but rejects the idea that doom is inevitable or even probable and says the odds are good.

Evidence for this placement

Signed the 2023 extinction statement [s7] and told the Senate that the risk of powerful AI threatening humanity as a whole is at least potentially real [s6]. In January 2026 he called misalignment a real risk with measurable probability while rejecting inevitability and saying the odds are good [s1]. In September 2026 he warned that recursive self-improvement could outrun our ability to control these systems [s3]. A serious, priority risk he believes is solvable: the +30 to +69 band.

Z · Regulatory intervention

+68
Minimal regulationStrong regulation
Medium confidence

Advocates binding federal regulation able to block unsafe models, plus a regulated, coordinated slowdown of frontier progress and attempts at international limits. He says pacing is not halting and doubts a full pause is achievable.

Evidence for this placement

In 2023 he asked the Senate for mandatory testing and auditing [s6]. In 2025 he favoured light-touch transparency laws [s5] and opposed a ten-year moratorium on state AI laws [s8]. In June 2026 he called for binding FAA-style regulation able to block releases [s2]; in September 2026 he said capability progress must be slowed through regulation, possible compute limits and international agreements [s3]. Top of the binding-rules band, bordering sweeping intervention; the position escalated during 2026.

“We must slow the pace at which we improve the capabilities of AI models.”
We Must Pace the Frontier · Dario Amodei (personal site) · 12 Sep 2026
Sources (9)
  1. [s1]
    The Adolescence of Technology
    Dario Amodei (personal site) · Jan 2026 · essay · primary source

    Long essay on the risks of powerful AI: autonomy, misuse, autocracy and labor-market disruption, with his preferred policy responses.

    darioamodei.com
  2. [s2]
    Policy on the AI Exponential
    Dario Amodei (personal site) · Jun 2026 · essay · primary source

    Policy essay saying it is time to move beyond transparency to binding, FAA-style regulation, with proposals on job displacement and taxes.

    darioamodei.com
  3. [s3]
    We Must Pace the Frontier
    Dario Amodei (personal site) · 12 Sep 2026 · essay · primary source

    Essay proposing embedded outside evaluators, coordinated pacing among democracies and attempts at global agreements to slow capability progress.

    darioamodei.com
  4. [s4]
    Machines of Loving Grace
    Dario Amodei (personal site) · Oct 2024 · essay · primary source

    Essay on the upside of powerful AI, with a closing section on work, meaning and the long-run economy.

    darioamodei.com
  5. [s5]
    The Urgency of Interpretability
    Dario Amodei (personal site) · Apr 2025 · essay · primary source

    Essay urging faster interpretability research and recommending light-touch transparency rules and chip export controls.

    darioamodei.com
  6. [s6]
    Written Testimony of Dario Amodei, hearing on “Oversight of A.I.: Principles for Regulation”
    U.S. Senate Committee on the Judiciary, Subcommittee on Privacy, Technology, and the Law · 25 Jul 2023 · testimony · primary source

    Prepared Senate testimony on short, medium and long-term AI risks, recommending a mandatory testing and auditing regime.

    judiciary.senate.gov
  7. [s7]
    Statement on AI Extinction Risk
    Center for AI Safety · May 2023 · open letter · primary source

    One-sentence open statement that mitigating extinction risk from AI should be a global priority; Amodei is listed among the signatories.

    safe.ai
  8. [s8]
    Anthropic CEO says proposed 10-year ban on state AI regulation 'too blunt' in NYT op-ed
    Reuters, via Yahoo News · 5 Jun 2025 · news report · secondary source

    Wire report on his New York Times op-ed opposing a moratorium on state AI laws and urging a federal transparency standard.

    yahoo.com
  9. [s9]
    Anthropic CEO Dario Amodei says AI industry needs to slow down for safety
    Associated Press, via Spectrum News 13 · 12 Sep 2026 · news report · secondary source

    Wire report on the pacing essay and the industry reaction; used here to date the essay to the day.

    mynews13.com
Editor’s notes and caveats

Z shifted in strength: mandatory testing (2023), light-touch transparency (2025), binding FAA-style rules (June 2026), then a regulated slowdown with possible compute limits and international agreements (September 2026). I rated Z medium because the rubric assigns that level to positions that have shifted; an editor could justify high, since two recent first-person essays agree. +68 sits on the boundary with the sweeping band; he explicitly says pacing is not halting and thinks a full pause is unlikely. The New York Times op-ed itself could not be loaded (paywall), so the moratorium point rests on Reuters. His essay pages give only month and year; the 12 September date for [s3] comes from the AP report [s9]. The essays cite events I did not examine in depth (an OpenAI and Hugging Face security incident, which PYMNTS also reports, and Claude Mythos Preview). Disclosure: this profile was researched by Claude, a model made by Anthropic, the company he runs, so a second human look at his placements is advisable.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Demis Hassabis

Chair of Google DeepMind and Chief Scientist of Alphabet; CEO of Isomorphic Labs

Editorial estimate

Hassabis believes AGI is a few years away and could bring radical abundance and cures for disease. A self-described cautious optimist, he worries about keeping control of autonomous systems and wants a standards body, eventually international, to test frontier models before release.

“we will need robust safeguards to maintain control of increasingly agentic, recursively self-improving systems”
A Framework for Frontier AI and the Dawning of a New Age · Demis Hassabis (Substack) · 14 Jul 2026

X · Economic impact

+15
Human complementarityHuman replacement
Medium confidence

Expects near-term tools that boost productivity and create new jobs, but says it may be different this time and that a post-AGI world of abundance will need new economic models.

Evidence for this placement

Asked by Wired in 2025 about job loss, he said new and better jobs usually appear, added that we will see whether this time is different, and said post-AGI abundance requires new economic theories [s2]. Asked by the Guardian whether abundance means mass unemployment, he called it one of the biggest things to figure out [s4]. His 2026 essay asks what economic models a post-scarcity world will need [s1]. He leaves the long-run outcome open: the explicitly mixed band, tilted slightly toward replacement.

Y · Existential risk

+40
Manageable riskExistential threat
High confidence

Counts loss of control over increasingly autonomous, self-improving systems among his main worries and wants it prioritised, while expressing confidence the technical risks can be solved if given time.

Evidence for this placement

Signed the 2023 extinction-risk statement [s7]. On 60 Minutes he named keeping control of more autonomous systems as one of his two main worries [s3]. He told Wired in 2025 that existential risk from today's systems is still theoretical [s2]. His July 2026 essay says safeguards will be needed to keep control of recursively self-improving systems and that the risks can be addressed given time [s1]. Serious, priority, solvable: the +30 to +69 band.

Z · Regulatory intervention

+55
Minimal regulationStrong regulation
High confidence

Wants a standards body that tests frontier models, with passing eventually required for deployment, growing into international standards. He has long backed IAEA-style oversight and would support a coordinated slowdown.

Evidence for this placement

In 2023 he proposed oversight starting like the IPCC and leading to a CERN for safety and an IAEA-style auditor [s5]. In 2025 he backed smart, nimble, international regulation [s2]. His July 2026 essay proposes a US standards body whose tests frontier models would eventually have to pass, able to coordinate a slowdown [s1]. He reportedly said he would back a universal pause [s6] and that Amodei's pacing essay points the right way [s8]. Binding frontier rules plus a regulator: the +30 to +69 band.

“Once you get to the post-AGI stage of radical abundance, new economic theories are required.”
Demis Hassabis Embraces the Future of Work in the Age of AI · Wired · 4 Jun 2025
Sources (9)
  1. [s1]
    A Framework for Frontier AI and the Dawning of a New Age
    Demis Hassabis (Substack) · 14 Jul 2026 · essay · primary source

    First-person essay proposing a US-led frontier AI standards body, initially voluntary and later mandatory, as a basis for international standards.

    demishassabis.substack.com
  2. [s2]
    Demis Hassabis Embraces the Future of Work in the Age of AI
    Wired · 4 Jun 2025 · interview · primary source

    Question-and-answer interview with Steven Levy covering AGI timelines, risk, regulation, jobs and post-AGI economics.

    wired.com
  3. [s3]
    Artificial intelligence could end disease, lead to "radical abundance," Google DeepMind CEO Demis Hassabis says
    CBS News, 60 Minutes · 20 Apr 2025 · interview · primary source

    Official transcript of Scott Pelley's 60 Minutes segment; first published April 2025, page updated August 2025.

    cbsnews.com
  4. [s4]
    Demis Hassabis on our AI future: ‘It’ll be 10 times bigger than the Industrial Revolution – and maybe 10 times faster’
    The Guardian · 4 Aug 2025 · news report · secondary source

    Interview-based profile quoting him on radical abundance, fair distribution, and being a cautious optimist.

    theguardian.com
  5. [s5]
    AI risk must be treated as seriously as climate crisis, says Google DeepMind chief
    The Guardian · 24 Oct 2023 · news report · secondary source

    News report of an interview before the UK AI Safety Summit, quoting his IPCC, CERN and IAEA proposals.

    theguardian.com
  6. [s6]
    AI CEOs want to slow down. The world’s too busy to help
    Transformer · 23 Jan 2026 · news report · secondary source

    Newsletter report from Davos 2026 quoting his preference for a slower pace and conditional support for a pause.

    transformernews.ai
  7. [s7]
    Statement on AI Extinction Risk
    Center for AI Safety · May 2023 · open letter · primary source

    One-sentence open statement that mitigating extinction risk from AI should be a global priority; Hassabis is listed among the signatories.

    safe.ai
  8. [s8]
    Tech leaders and governments split over ‘AI doom’ fears
    Reuters, via TimesLIVE · 21 Sep 2026 · news report · secondary source

    Wire roundup quoting his response to Amodei's pacing essay and noting his move to Alphabet chief scientist.

    timeslive.co.za
  9. [s9]
    Demis Hassabis no longer DeepMind CEO to focus on AGI role
    9to5Google · 5 Aug 2026 · news report · secondary source

    Report on Google's announcement that he becomes Chair of Google DeepMind and Chief Scientist of Alphabet, remaining CEO of Isomorphic Labs.

    9to5google.com
Editor’s notes and caveats

Role changed on 5 August 2026 (per [s9] and [s8]); I did not load a Google primary announcement. The Substack essay [s1] predates the change and is signed as CEO of Google DeepMind. X is thinner than Y and Z: he consistently gives a near-term augmentation answer while saying post-AGI economics is unresolved, so +15 is a judgement within the mixed band. The Davos pause remark [s6] and the September 2026 reaction [s8] are from reporting; the World Economic Forum transcript refused automated requests. Wired's page displays 4 June 2025, while its metadata gives 3 June as the publication date. The Nobel Prize interview page I loaded contained nothing relevant to these axes. His concrete 2026 proposal is an industry-funded, federally overseen body that starts voluntary, which is milder than the IAEA framing; an editor could place Z a few points lower.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Eliezer Yudkowsky

Co-founder, Machine Intelligence Research Institute (MIRI)

Editorial estimate

Believes that building superhuman AI with anything like today's methods would kill everyone, and that nobody is close to making it safe. Wants an indefinite, enforced worldwide halt to frontier AI development. Treats ordinary job loss as minor next to extinction.

“If we go ahead on this everyone will die”
Pausing AI Developments Isn't Enough. We Need to Shut it All Down · TIME (full text republished by the Machine Intelligence Research Institute) · 29 Mar 2023

X · Economic impact

+80
Human complementarityHuman replacement
Medium confidence

Expects automation before superintelligence to cause ordinary churn rather than lasting unemployment, but thinks a superintelligence would do everything better and have no economic use for humans at all.

Evidence for this placement

In a 2024 essay he argues comparative advantage does not protect humans from a superintelligence, noting that horses went to glue factories after cars [s3]. MIRI's 2025 position paper, which he co-authored, says keeping humans around is unlikely to be the efficient solution to any problem an AI has [s2]. A December 2023 post separates temporary 'disemployment' from lasting unemployment and calls job loss before AGI the standard cost of progress [s4]. Long-run view is total substitution, the top band; medium because his remarks concern survival, not labor markets.

Y · Existential risk

+98
Manageable riskExistential threat
High confidence

Thinks that building superhuman AI under anything like present conditions leads, by default, to the death of every human, and that the alignment problem is nowhere near solved.

Evidence for this placement

His 2023 TIME op-ed says the most likely result of building superhumanly smart AI under current circumstances is that literally everyone on Earth dies [s1]. MIRI's 2025 position paper, which he co-authored, says the expected outcome of superintelligence built with current methods is human extinction [s2]. His 2025 book with Nate Soares is titled 'If Anyone Builds It, Everyone Dies' [s5]. Repeated, explicit and consistent over years, so placed at the far end of the catastrophe-by-default band.

Z · Regulatory intervention

+98
Minimal regulationStrong regulation
High confidence

Calls for an indefinite, worldwide moratorium on large AI training runs, enforced through international agreement, hardware tracking and falling compute caps, with no exemptions for governments or militaries.

Evidence for this placement

The TIME op-ed asks governments to shut down large GPU clusters and training runs, cap and then lower permitted compute, track all GPUs sold, and be willing to destroy a rogue datacenter by airstrike under an international agreement [s1]. MIRI's 2025 paper, which he co-authored, calls for an international 'off switch' able to shut down or ban frontier AI development [s2]. International agreements, moratoria and compute limits together define the top band of the rubric.

“The moratorium on new large training runs needs to be indefinite and worldwide.”
Pausing AI Developments Isn't Enough. We Need to Shut it All Down · TIME (full text republished by the Machine Intelligence Research Institute) · 29 Mar 2023
Sources (5)
  1. [s1]
    Pausing AI Developments Isn't Enough. We Need to Shut it All Down
    TIME (full text republished by the Machine Intelligence Research Institute) · 29 Mar 2023 · essay · primary source

    Op-ed first published in TIME on March 29, 2023. This is MIRI's own full-text copy, posted April 7 with an author's addendum.

    intelligence.org
  2. [s2]
    The Problem
    LessWrong (crosspost of MIRI's position paper) · 5 Aug 2025 · essay · primary source

    MIRI's statement of its position, with Yudkowsky, Soares and four colleagues on the byline. First posted on intelligence.org in February 2025.

    lesswrong.com
  3. [s3]
    The Sun is big, but superintelligences will not spare Earth a little sunlight
    LessWrong · 23 Sep 2024 · essay · primary source

    Essay crossposted from his X account with his permission, rebutting claims that trade or comparative advantage would lead a superintelligence to spare humanity.

    lesswrong.com
  4. [s4]
    X post on compensating people who lose work to AI ('disemployment')
    X (formerly Twitter) · 29 Dec 2023 · social post · primary source

    Long post arguing that job loss from pre-AGI systems is the normal cost of progress, unlike AI later killing everyone. Timestamp is UTC.

    x.com
  5. [s5]
    "If Anyone Builds It, Everyone Dies" release day!
    Machine Intelligence Research Institute · 16 Sep 2025 · news report · secondary source

    MIRI staff announcement of the publication of Yudkowsky and Soares's book. Cited only for the book's existence, title and authorship.

    intelligence.org
Editor’s notes and caveats

Positions on Y and Z are stable from 2023 to 2025. X is the weakest coordinate: he rarely discusses labor markets as such, and his long-run forecast is extinction, not a post-work economy; an editor could reasonably leave X as low confidence. time.com blocked automated access (HTTP 406), so the op-ed was verified on MIRI's republication. The book's companion site ifanyonebuildsit.com sits behind a bot-check page and was not read; nothing here rests on it, and the book text itself was not consulted. [s4] is a long X post: x.com serves only a JavaScript shell, so the opening lines were confirmed through X's own embed endpoint (cdn.syndication.twimg.com) and the remainder, including the 'disemployment' passage, was read through the third-party FxTwitter API mirror. A verified alternate quotation for X, from [s3]: It's not always profitable to pay horses enough hay for them to live on. Title 'Co-Founder' is as shown on intelligence.org/team on 2026-10-13. In [s1] he says he declined to sign the 2023 FLI pause letter because it asked for too little.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Emad Mostaque

Founder and CEO, Intelligent Internet; founder and former CEO of Stability AI

Editorial estimate

Mostaque, who led Stability AI, believes AI will make human cognitive labor economically worthless within years and rates the chance of catastrophe a coin toss. He thinks regulation is too late and wants open-source AI provided to every citizen as public infrastructure instead.

“My P(Doom) is fifty percent.”
Emad Mostaque Has A 50% P(Doom) & A Plan To Lower It · Doom Debates (Liron Shapira) · 21 Apr 2026

X · Economic impact

+90
Human complementarityHuman replacement
High confidence

Predicts AI will out-compete human minds across cognitive work, leaving people economically irrelevant with nowhere left to pivot, and that the decisive transition is roughly a thousand days away.

Evidence for this placement

His book The Last Economy says AI replaces the mind itself, that for a growing majority of cognitive tasks a human's economic value is negative, and that there is nowhere left to pivot [s1][s2]. In April 2026 he said that even without catastrophe people's jobs are screwed [s3]; in 2024 he called the Filipino call-centre sector dead [s4]. He wants the outcome reshaped, but the prediction itself is near-total substitution: the top band.

Y · Existential risk

+68
Manageable riskExistential threat
High confidence

Puts the chance of AI catastrophe at fifty percent, describes the future as either doom or utopia, and doubts even his own plan could cut the risk below thirty percent.

Evidence for this placement

Signed both the 2023 pause letter [s5] and the extinction-risk statement [s6], and told the BBC the worst case is AI that controls humanity [s7]. In an April 2026 interview he said his P(Doom) is fifty percent, called anyone claiming zero a liar, and said he struggles to see it falling below thirty percent [s3]. A coin toss is just short of catastrophe by default, so this sits at the top of the serious-risk band.

Z · Regulatory intervention

−15
Minimal regulationStrong regulation
Medium confidence

Thinks regulation and treaties can no longer stop frontier AI. He backs narrow rules, such as data transparency for decision-making AI and strict limits on voice cloning, and rejects liability and compute caps.

Evidence for this placement

He signed the 2023 pause letter [s5], but by December 2024 said it was too late, that governments cannot afford to regulate, that liability will not work and compute caps are impossible, while recommending data-transparency rules and heavy regulation of speech AI [s4]. In April 2026 he said a pause treaty would be valuable but will not happen and that regulation will not stop the most powerful AIs [s3]. His book prefers incentives to mandates [s8]. Targeted rules, heavier mandates rejected, record mixed: the -29 to +29 band.

“In the long run, we're all alive, but economically irrelevant.”
The Last Economy: A Guide to the Age of Intelligent Economics — Introduction: The Thousand-Day Window · Intelligent Internet (ii.inc) · 2025
Sources (8)
  1. [s1]
    The Last Economy: A Guide to the Age of Intelligent Economics — Introduction: The Thousand-Day Window
    Intelligent Internet (ii.inc) · 2025 · book · primary source

    Introduction to his book, published free online by his company; the page says it was last updated on 8 October 2026.

    ii.inc
  2. [s2]
    The Last Economy — Chapter 1: The Intelligence Inversion
    Intelligent Internet (ii.inc) · 2025 · book · primary source

    Book chapter arguing AI is a final inversion that leaves human cognitive labor with no economic refuge.

    ii.inc
  3. [s3]
    Emad Mostaque Has A 50% P(Doom) & A Plan To Lower It
    Doom Debates (Liron Shapira) · 21 Apr 2026 · podcast · primary source

    Long podcast interview with a published transcript, covering his risk estimate, the economy, and whether a pause treaty could work.

    lironshapira.substack.com
  4. [s4]
    Emad Mostaque on the Intelligent Internet and Universal Basic AI
    The Cognitive Revolution · 25 Dec 2024 · podcast · primary source

    Podcast interview with an automated transcript; includes direct questions on the pause letter and which regulations he would recommend.

    cognitiverevolution.ai
  5. [s5]
    Pause Giant AI Experiments: An Open Letter
    Future of Life Institute · 22 Mar 2023 · open letter · primary source

    Open letter calling for a six-month pause on training systems more powerful than GPT-4; he is listed as CEO, Stability AI.

    futureoflife.org
  6. [s6]
    Statement on AI Extinction Risk
    Center for AI Safety · May 2023 · open letter · primary source

    One-sentence open statement that mitigating extinction risk from AI should be a global priority; Mostaque is listed among the signatories.

    safe.ai
  7. [s7]
    A.I. ‘controls humanity’ in the worst-case scenario but will probably just find us boring, says Stability AI CEO Emad Mostaque
    Fortune · 14 May 2023 · news report · secondary source

    Report on a BBC interview given while he was Stability AI's chief executive, quoting his worst-case scenario.

    fortune.com
  8. [s8]
    The Last Economy — Chapter 17: The Symbiotic State
    Intelligent Internet (ii.inc) · 2025 · book · primary source

    Book chapter describing the state as a gardener that uses incentives and public infrastructure, not mandates.

    ii.inc
Editor’s notes and caveats

Z is the hardest axis: he signed a moratorium letter in 2023 and still calls a pause treaty valuable in principle, but since late 2024 has said regulation is too late and unworkable, and his programme is public provision of open-source AI, not rules. The rubric scores regulation advocated, hence a slightly negative mixed placement; a reader could argue for anything from -30 to +10. Y at +68 treats a stated coin toss as just below catastrophe by default; +70 is defensible. The Cognitive Revolution transcript [s4] is machine-generated with visible errors, so I used it for substance only and took no quotation from it. His name on the pause letter was confirmed in the signatory data file the FLI page loads, not in the static page. The book's publication month (reported as August 2025) comes from secondary sources, and the online text was updated in October 2026, so wording may differ from the first edition. His 2023 letter to the US Senate on open models could not be loaded and is not cited. The Center for AI Safety page shows no date; May 2023 is from general knowledge of the statement's release.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Fei-Fei Li

Co-founder and CEO, World Labs; Sequoia Professor of Computer Science and founding co-director of the Institute for Human-Centered AI, Stanford University

Editorial estimate

Stanford computer scientist behind ImageNet and co-founder of World Labs. A self-described pragmatic optimist who wants AI to augment people, rejects both utopian and doomsday narratives, and favors evidence-based rules such as transparency over mandates that would burden open research.

“in terms of urgency, I'm more concerned about ameliorating the risks that are here and now”
AI pioneer Fei-Fei Li: 'I'm more concerned about the risks that are here and now' · The Guardian (The Observer) · 5 Nov 2023

X · Economic impact

−30
Human complementarityHuman replacement
Medium confidence

Calls predictions that everyone's job will disappear hyperbolic and says AI should augment people. Also expects painful labor-market change whose scale she thinks is underappreciated, with outcomes depending on human choices.

Evidence for this placement

On Tim Ferriss's show (December 2025) she called the rhetoric that 'everybody's job will be gone' hyperbolic, alongside post-scarcity utopia, and said AI's labor-market impact is underappreciated [s4]. In a 2024 interview she said AI 'can enhance or take away human jobs' and named 'painful labor market change' as a risk [s5]. Her 2025 essay says AI 'must augment human capability, not replace it', which is a goal rather than a forecast [s3]. Leans to augmentation but treats the outcome as partly open: boundary with the mixed band.

Y · Existential risk

−45
Manageable riskExistential threat
High confidence

Respects the existential concern and does not call it silly, but says it is overhyped and not her priority. Focuses on present harms and treats apocalypse narratives as an extreme.

Evidence for this placement

Asked by the Guardian (November 2023) about AI destroying humanity, she said 'I respect the existential concern' but worries more about risks 'here and now' [s1]; she told MIT Technology Review other risks are 'more pressing and urgent' [s8]. A 2024 report quotes her calling extinction risk overhyped and 'blown out of control' [s10]. Her 2025 essay sets aside 'narratives of techno-utopia and apocalypse' as extreme [s3]. The California report she co-led lists loss of control among risks with growing evidence [s6]. Remote, not dismissed: second band.

Z · Regulatory intervention

+5
Minimal regulationStrong regulation
High confidence

Backs evidence-based, targeted rules such as transparency, third-party evaluation and incident reporting. Opposed California's SB 1047 over developer liability, a kill-switch mandate and compute thresholds, warning of harm to open source and academia.

Evidence for this placement

Her August 2024 Fortune op-ed opposed SB 1047's developer liability and kill switch but said 'I am not anti-AI governance' [s2]. She then co-led California's frontier AI working group, whose June 2025 report recommends transparency, third-party assessment, whistleblower protection and adverse-event reporting in a 'trust but verify' spirit [s6], and co-signed a statement welcoming the SB 53 transparency law [s7]. In February 2025 she urged policy grounded in 'science, not science fiction' [s9]. Targeted rules yes, heavier mandates no: middle band, tilted by her recent support for binding transparency.

“I am not anti-AI governance. Legislation is critical to the safe and effective advancement of AI.”
'The Godmother of AI' says California's well-intended AI bill will harm the U.S. ecosystem · Fortune · 6 Aug 2024
Sources (10)
  1. [s1]
    AI pioneer Fei-Fei Li: 'I'm more concerned about the risks that are here and now'
    The Guardian (The Observer) · 5 Nov 2023 · interview · primary source

    Question-and-answer interview with Zoë Corbyn marking the publication of her memoir, days after the UK AI Safety Summit.

    theguardian.com
  2. [s2]
    'The Godmother of AI' says California's well-intended AI bill will harm the U.S. ecosystem
    Fortune · 6 Aug 2024 · essay · primary source

    Her bylined commentary opposing California's SB 1047 while the bill was before the state legislature.

    fortune.com
  3. [s3]
    From Words to Worlds: Spatial Intelligence is AI's Next Frontier
    Fei-Fei Li on Substack · 10 Nov 2025 · essay · primary source

    Her own essay on spatial intelligence and world models, with a closing section on what should motivate AI development.

    drfeifei.substack.com
  4. [s4]
    The Tim Ferriss Show Transcripts: Dr. Fei-Fei Li, The Godmother of AI — Asking Audacious Questions, Civilizational Technology, and Finding Your North Star (#839)
    The Tim Ferriss Show (tim.blog) · 10 Dec 2025 · podcast · primary source

    Host-published transcript of a long interview; includes her remarks on public worry about AI, the labor market and hiring.

    tim.blog
  5. [s5]
    "AI Is a Tool, and Its Values Are Human Values."
    Issues in Science and Technology · 9 Apr 2024 · interview · primary source

    Question-and-answer interview in the Spring 2024 issue on human-centered AI, policy and the National AI Research Resource.

    issues.org
  6. [s6]
    The California Report on Frontier AI Policy
    Joint California Policy Working Group on AI Frontier Models (arXiv copy) · 17 Jun 2025 · paper · primary source

    Policy report commissioned by Governor Newsom after the SB 1047 veto; Li is one of three co-leads.

    arxiv.org
  7. [s7]
    Governor Newsom signs SB 53, advancing California's world-leading artificial intelligence industry
    Office of the Governor of California · 29 Sep 2025 · news report · primary source

    Official press release carrying a joint statement by Cuéllar, Li and Chayes saying the law advances their report's transparency principles.

    gov.ca.gov
  8. [s8]
    AI is at an inflection point, Fei-Fei Li says
    MIT Technology Review · 14 Nov 2023 · news report · secondary source

    Newsletter write-up by Melissa Heikkilä of an interview with Li, with direct quotes on catastrophic versus existential risk.

    technologyreview.com
  9. [s9]
    AI pioneer Fei-Fei Li says AI policy must be based on 'science, not science fiction'
    TechCrunch · 8 Feb 2025 · news report · secondary source

    Brief report quoting the three policymaking principles from her Financial Times op-ed ahead of the Paris AI Action Summit.

    techcrunch.com
  10. [s10]
    AI Doomsday or Panacea? A Leading Voice Weighs In
    The AI Insider · 15 May 2024 · news report · secondary source

    Report quoting her on-stage remarks about extinction risk at the Bloomberg Technology Summit in May 2024.

    theaiinsider.tech
Editor’s notes and caveats

X is the weakest axis. Most of what she says about work is about what AI should do (augment people), not a forecast; the placement rests on a few hedged first-person remarks and sits on the boundary between the augmentation and mixed bands, so -20 to -40 are all defensible. Sources I could not use: her November 2025 Lenny's Podcast episode on jobs (transcript paywalled) and a June 2026 Silicon Valley Girl podcast episode (found only as an unlabelled machine transcript with a second guest, so attribution was unsafe). Her February 2025 Financial Times op-ed is paywalled, so its wording is cited through TechCrunch [s9]; the Bloomberg Technology Summit remark is cited through a secondary report [s10]. On Z, note the sequence: opposition to SB 1047 in 2024, then support for transparency rules in 2025. The California report states that it does not argue for or against any particular legislation, and [s7] is a joint statement rather than her words alone. Her memoir 'The Worlds I See' was not consulted directly. The CEO title was confirmed in the introduction to [s4].

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Gary Marcus

Professor Emeritus of Psychology and Neural Science, New York University; author of the newsletter Marcus on AI

Editorial estimate

Cognitive scientist and persistent critic of AI hype. He doubts today's systems are close to human-level intelligence and calls near-term extinction fears preposterous, yet campaigns for a dedicated AI regulator, pre-release safety review and company liability.

“The idea that AI will kill us all in five years is preposterous.”
No, Anderson Cooper, AI is not going to kill all humans by 2030 · Marcus on AI (Substack) · 10 Sep 2026

X · Economic impact

+35
Human complementarityHuman replacement
Medium confidence

Says today's unreliable AI will mostly assist workers and that mass layoffs are not imminent. But he expects true general intelligence, perhaps decades away, eventually to replace a large fraction of human jobs.

Evidence for this placement

Senate testimony (May 2023): this technology will differ from earlier ones, and in the long run AGI 'really will replace a large fraction of human jobs', on a timescale he put anywhere from 10 to 100 years [s1]. His April 2026 Fortune essay argues current AI will more often augment jobs, with real change likely only after advances 'a decade or more away' [s7][s10]. Long-run substitution, heavy near-term scepticism: low end of the +30 to +69 band.

Y · Existential risk

−50
Manageable riskExistential threat
High confidence

Considers literal human extinction from AI extremely unlikely and near-term takeover scenarios preposterous. He keeps this separate from catastrophes short of extinction, such as cyberattacks or bioweapons, which he takes seriously.

Evidence for this placement

August 2023 essay: not very worried about extinction, though the chance is not zero [s8]. July 2025: put his p(doom) at 'maybe 3%' after the Grok 4 launch, still below most industry estimates [s3]. September 2026: existential risk is 'barely above zero', AI killing everyone within five years is 'preposterous', while catastrophic risk is 'quite real' [s2]; he said the same at a UN event [s5]. Unlikely and remote, focus on concrete harms: the -69 to -30 band.

Z · Regulatory intervention

+68
Minimal regulationStrong regulation
High confidence

Wants a dedicated AI agency, FDA-style safety review before wide release, auditing, company liability and international agreements. He is open to a temporary pause but opposes a permanent, unilateral ban on superhuman AI research.

Evidence for this placement

Senate testimony (May 2023): pre-deployment safety review 'like we use with the FDA', a monitoring agency able to recall systems, and a CERN-like international body [s1]. He defended SB 1047 as modest [s9] and in 2025 called for liability, auditing and international treaties [s3]. In September 2026 he argued for regulation alongside liability [s6] and an international pre-release commission [s5], but opposed the Sanders-Casar superintelligence ban as too broad [s4]. Top of the binding-rules band.

“a safety review like we use with the FDA prior to widespread deployment”
Oversight of A.I.: Rules for Artificial Intelligence (Senate Hearing 118-037) · U.S. Government Publishing Office · 16 May 2023
Sources (10)
  1. [s1]
    Oversight of A.I.: Rules for Artificial Intelligence (Senate Hearing 118-037)
    U.S. Government Publishing Office · 16 May 2023 · testimony · primary source

    Official record of the Senate Judiciary subcommittee hearing at which he testified alongside Sam Altman and Christina Montgomery.

    govinfo.gov
  2. [s2]
    No, Anderson Cooper, AI is not going to kill all humans by 2030
    Marcus on AI (Substack) · 10 Sep 2026 · essay · primary source

    Newsletter essay rejecting near-term extinction claims and distinguishing existential from catastrophic risk.

    garymarcus.substack.com
  3. [s3]
    Why my p(doom) has risen, dramatically
    Marcus on AI (Substack) · 15 Jul 2025 · essay · primary source

    Newsletter essay written after the Grok 4 launch, giving a rough extinction probability and calling for serious regulation.

    garymarcus.substack.com
  4. [s4]
    The new Sanders-Casar Ban Artificial Superintelligence Act – and why I oppose it
    Marcus on AI (Substack) · 3 Sep 2026 · essay · primary source

    Newsletter post explaining why he opposes a proposed permanent US ban while still supporting an AI agency and a possible pause.

    garymarcus.substack.com
  5. [s5]
    Big news at the UN
    Marcus on AI (Substack) · 21 Sep 2026 · talk · primary source

    Newsletter post reproducing the framing remarks he gave at a UN General Assembly digital cooperation event.

    garymarcus.substack.com
  6. [s6]
    Liability, regulation, and AI's new false dichotomy
    Marcus on AI (Substack) · 17 Sep 2026 · essay · primary source

    Newsletter essay arguing that liability and regulation are both needed, with reference to his 2023 exchange with Senator Hawley.

    garymarcus.substack.com
  7. [s7]
    9 reasons AI isn't going to take your job (yet)
    Fortune · 1 Apr 2026 · essay · primary source

    His bylined magazine commentary advising employers not to rush into replacing staff with AI.

    fortune.com
  8. [s8]
    p(doom)
    Marcus on AI (Substack) · 27 Aug 2023 · essay · primary source

    Newsletter essay on why he doubts near-term extinction but worries about lesser catastrophes.

    garymarcus.substack.com
  9. [s9]
    The misguided backlash against California's SB-1047
    Marcus on AI (Substack) · 7 Jun 2024 · essay · primary source

    Newsletter essay defending California's frontier AI safety bill against industry criticism.

    garymarcus.substack.com
  10. [s10]
    On employment, don't panic – yet.
    Marcus on AI (Substack) · 1 Apr 2026 · essay · primary source

    Short newsletter post summarising his Fortune commentary for employees.

    garymarcus.substack.com
Editor’s notes and caveats

Y and Z point in opposite directions and are scored separately. Y counts only extinction or loss of control; his serious concern about catastrophes short of extinction does not raise it. His July 2025 figure of about 3 percent was tied to one scenario, and by September 2026 he was back to 'barely above zero'. X is the least certain: his recent writing, read alone, would land in the augmentation range, and the placement follows the rubric's long-run framing plus his 2023 testimony, which he has echoed since only loosely ('probably someday it will'). An editor could reasonably choose anything from +15 to +40. Z: he told the Senate he signed the March 2023 pause letter 'spiritually' and favoured pausing deployment rather than research [s1]. I could not confirm whether he signed the 2023 CAIS statement because safe.ai serves only the first 50 names in its static page. His books Rebooting AI and Taming Silicon Valley were not consulted directly. Several 2026 posts refer to events I know only from his posts and the CNN transcript used for Suleyman.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Geoffrey Hinton

Emeritus Professor of Computer Science, University of Toronto; 2024 Nobel laureate in Physics

Editorial estimate

A founder of deep learning who left Google in 2023 to speak freely about its dangers. He expects AI to outthink people, replace most intellectual work and possibly escape human control, and he wants governments to regulate developers firmly.

“We have no idea whether we can stay in control.”
Geoffrey Hinton: Banquet speech · NobelPrize.org · 10 Dec 2024

X · Economic impact

+75
Human complementarityHuman replacement
High confidence

Expects AI to replace nearly everyone doing routine intellectual work and, once superintelligent, to outperform people at everything. He doubts new jobs will make up the loss, though physical trades and healthcare should last longer.

Evidence for this placement

On Diary of a CEO (June 2025) he said AI will 'replace everybody' in mundane intellectual labor and that with superintelligence 'nothing remains' [s3]. On CNN (Dec 2025) he said AI will make intelligence 'more or less irrelevant' [s4]. In Geneva (April 2026), per AFP, he said superintelligent AI could do any intellectual job [s8]. He exempts plumbing and healthcare for now, so this sits at the bottom of the top band.

Y · Existential risk

+60
Manageable riskExistential threat
High confidence

Sees a real chance, which he puts at roughly 10 to 20 percent, that AI smarter than us takes control. He calls both certain doom and certain safety unreasonable and wants urgent research on staying in control.

Evidence for this placement

Nobel banquet speech (Dec 2024): a 'longer term existential threat', with no idea whether we can stay in control [s1]. Nobel interview: people sure of takeover and people sure of safety are both 'crazy' [s2]. On CNN (Dec 2025) he confirmed his 10 to 20 percent takeover estimate as 'a very real fear' [s4]. He signed the 2023 extinction-risk statement [s6]. Serious and prioritised but explicitly uncertain: upper part of the +30 to +69 band.

Z · Regulatory intervention

+72
Minimal regulationStrong regulation
High confidence

Calls for strong, binding regulation: mandatory safety testing for large developers and forceful action by governments and international bodies. In 2025 he also signed a call to prohibit building superintelligence until it is shown to be safe.

Evidence for this placement

Nobel banquet speech: AI risks need 'urgent and forceful attention from governments and international organizations' [s1]. Diary of a CEO: 'we need strong regulation' [s3]. CNN: companies 'should be required to do significant testing' [s4]. He co-signed the 2024 letter backing California's SB 1047 [s7] and the 2025 Statement on Superintelligence, which calls for a prohibition [s5]. That signature puts him just inside the top band; his spoken proposals are mostly binding rules.

“for mundane intellectual labor, AI is just going to replace everybody”
Godfather of AI: I Tried to Warn Them, But We've Already Lost Control! Geoffrey Hinton (transcript) · The Diary Of A CEO with Steven Bartlett, transcript hosted by PodScripts · 16 Jun 2025
Sources (9)
  1. [s1]
    Geoffrey Hinton: Banquet speech
    NobelPrize.org · 10 Dec 2024 · talk · primary source

    His speech at the Nobel Prize banquet in Stockholm, published in full by the Nobel Foundation.

    nobelprize.org
  2. [s2]
    Transcript from an interview with Geoffrey Hinton
    NobelPrize.org · 6 Dec 2024 · interview · primary source

    Official transcript of the laureate interview recorded during Nobel Week in Stockholm.

    nobelprize.org
  3. [s3]
    Godfather of AI: I Tried to Warn Them, But We've Already Lost Control! Geoffrey Hinton (transcript)
    The Diary Of A CEO with Steven Bartlett, transcript hosted by PodScripts · 16 Jun 2025 · podcast · primary source

    Long-form podcast interview. The page is an unofficial machine transcript of the episode, not one published by the show.

    podscripts.co
  4. [s4]
    State of the Union: Interview With Geoffrey Hinton (rush transcript)
    CNN · 28 Dec 2025 · interview · primary source

    CNN's own transcript of Jake Tapper's interview in a special episode on AI. Marked by CNN as a rush transcript.

    transcripts.cnn.com
  5. [s5]
    Statement on Superintelligence
    Future of Life Institute · Oct 2025 · open letter · primary source

    One-sentence public statement calling for a prohibition on developing superintelligence. The page's signatory data lists Hinton as a verified signer.

    superintelligence-statement.org
  6. [s6]
    Statement on AI Extinction Risk
    Center for AI Safety · 30 May 2023 · open letter · primary source

    One-sentence statement that extinction risk from AI should be a global priority. Hinton is the first listed signatory.

    safe.ai
  7. [s7]
    Exclusive: Renowned Experts Pen Support for California's Landmark AI Safety Bill
    TIME · 7 Aug 2024 · news report · secondary source

    News report quoting a letter from Bengio, Hinton, Lessig and Russell urging California leaders to pass SB 1047.

    time.com
  8. [s8]
    AI godfather Hinton: Unregulated AI like speeding with no steering wheel
    AFP via The Star · 22 Apr 2026 · news report · secondary source

    Wire report on his video address to the 2026 Digital World Conference in Geneva, covering regulation, jobs and coexistence with superintelligent AI.

    thestar.com.my
  9. [s9]
    The Godfather of AI reveals which jobs are safest — and where 'everybody' will get replaced
    Business Insider via AOL · 16 Jun 2025 · news report · secondary source

    News report on the Diary of a CEO interview, used here to cross-check the wording of the transcript.

    aol.com
Editor’s notes and caveats

X: he separates the near term (call centres, paralegals, most routine office work) from the superintelligence case where nothing is left; an editor reading the axis more narrowly could justify +60 to +65. Z: the +72 depends on his signature on the superintelligence prohibition statement; without it his own stated proposals (required testing, strong regulation) would score in the 50s or 60s. The SB 1047 letter itself could not be loaded (safesecureai.org did not respond), so I relied on TIME's report of it. The Statement on Superintelligence date is given to the month: signature records in the page data are stamped October 2025, while FLI's press page carries a 27 March 2026 publish stamp. The podcast transcript in s3 is unofficial; the quoted sentence was cross-checked against Business Insider. His November 2025 Georgetown event with Senator Sanders appeared only in search summaries and was not used.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Jensen Huang

Founder and CEO, NVIDIA

Editorial estimate

Nvidia's chief executive calls AI essential infrastructure and a new industrial revolution that will create jobs and raise productivity. He dismisses extinction warnings as unscientific doom narratives, says existing laws are enough, and opposes slowing development or banning chip sales to China.

“The number of software engineers at NVIDIA is gonna grow, not decline.”
Transcript for Jensen Huang: NVIDIA – The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494 · Lex Fridman · 23 Mar 2026

X · Economic impact

−80
Human complementarityHuman replacement
High confidence

Insists AI changes the tasks within a job, not its purpose, so productivity gains lead employers to hire more. Concedes every job will change and some will vanish, but calls job-apocalypse forecasts wrong.

Evidence for this placement

His March 2026 essay [s1] says AI makes hospitals more productive so they hire more people, and that the AI buildout needs enormous skilled labor. On Lex Fridman [s2] he said Nvidia's software engineers will grow in number; on Dwarkesh [s3] he attacked predictions of the end of work, citing a radiologist shortage. In 2025 he said some jobs will be obsolete but many created [s6], and that people will lose jobs to someone using AI rather than to AI [s7]. Emphatic augmentation, rejecting mass-unemployment forecasts: -100 to -70 band.

Y · Existential risk

−85
Manageable riskExistential threat
High confidence

Rejects predictions of AI-caused extinction as made up and not grounded in science, and says frightening people is irresponsible. Treats loss-of-control incidents as engineering failures that labs can diagnose and fix.

Evidence for this placement

CBS News reported in September 2026 that he said there is '0% chance' the world ends in 2030 and that such predictions are not grounded in science [s4]. At the All-In Summit that month he called extinction probabilities made up and said recursive self-improvement will not spiral out of control because labs evaluate what they release, while calling safety paramount [s5]. On Dwarkesh he said leaving an AI agent unwatched would be insane and expects monitoring agents to keep AI safe [s3]. Dismissal as unfounded: -100 to -70 band.

Z · Regulatory intervention

−70
Minimal regulationStrong regulation
High confidence

Says no new AI-specific rules are needed: apply existing liability and cybersecurity law, solve safety through engineering and independent evaluators, and keep moving fast. Opposes slowdowns and bans on selling chips to China.

Evidence for this placement

CBS reported in September 2026 that he dismissed the need for new regulations, saying existing liability and cybersecurity laws should be applied first, and objected to bans on chip sales to China [s4]. At the All-In Summit he said regulation should solve actual problems, called pausing a voluntary choice for companies, and accepted multiple third-party evaluators [s5]. On Dwarkesh he argued at length against chip export controls [s3]. Opposes new AI-specific regulation but accepts existing law and audits: placed on the boundary between the -100 to -70 and light-touch bands.

“There is 0% chance that's going to be the end of the world”
Nvidia's Jensen Huang rejects AI extinction warnings as "doomsday narratives" · CBS News · 20 Sep 2026
Sources (7)
  1. [s1]
    AI Is a 5-Layer Cake
    NVIDIA Blog · 10 Mar 2026 · essay · primary source

    First-person essay bylined to Huang describing AI as infrastructure and its effects on labor and productivity.

    blogs.nvidia.com
  2. [s2]
    Transcript for Jensen Huang: NVIDIA – The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494
    Lex Fridman · 23 Mar 2026 · podcast · primary source

    Podcast's own human-generated, speaker-labeled transcript of a long interview; includes a section on jobs and programming.

    lexfridman.com
  3. [s3]
    Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat
    Dwarkesh Podcast · 15 Apr 2026 · podcast · primary source

    Podcast's own speaker-labeled transcript; Huang debates export controls, AI security and job-loss predictions with Dwarkesh Patel.

    dwarkesh.com
  4. [s4]
    Nvidia's Jensen Huang rejects AI extinction warnings as "doomsday narratives"
    CBS News · 20 Sep 2026 · interview · secondary source

    CBS News article on its own Sunday Morning interview, written by interviewer Jo Ling Kent; direct quotes, not a full transcript.

    cbsnews.com
  5. [s5]
    Jensen Huang: The Doomer Hoax, Superintelligence Is Here, and The Future of AI (ft. President Trump)
    All-In Podcast (transcript hosted by HappyScribe) · 14 Sep 2026 · podcast · primary source

    Unofficial machine transcript, without speaker labels, of his on-stage All-In Summit interview; speaker attribution inferred from context.

    podcasts.happyscribe.com
  6. [s6]
    Nvidia's Jensen Huang says he disagrees with almost everything Anthropic CEO Dario Amodei says
    Fortune · 11 Jun 2025 · news report · secondary source

    Report on a press briefing at VivaTech in Paris where he rejected Amodei's entry-level job-loss forecast.

    fortune.com
  7. [s7]
    Nvidia CEO: You won't lose your job to AI—you'll 'lose your job to somebody who uses AI'
    CNBC Make It via NBC Los Angeles · 28 May 2025 · news report · secondary source

    Syndicated CNBC report on his remarks at the Milken Institute Global Conference on May 6, 2025.

    nbclosangeles.com
Editor’s notes and caveats

All three placements are consistent across 2025-2026 sources; no shift found on X or Y. On Z, a tech-press piece (The Next Web, 19 Sept 2026, loaded but not cited) argued his CBS line about going as fast as possible sits in tension with his All-In remark days earlier that labs could pause or pace voluntarily if out of control; both are compatible with opposing mandated slowdowns. Caveats: [s5] is an unofficial machine transcript with no speaker labels, so it is used for evidence but not for quotations. [s4] is the interviewing outlet's article rather than a transcript; a 46-minute extended video on the same page was not reviewed. Earnings-call transcripts were not reachable in a verifiable form, so his export-control position is sourced to [s3] and [s4]. Search summaries (not loaded, so not cited) reported January 2026 No Priors remarks about a science-fiction doomer narrative and regulatory capture, 2023 remarks that AI needs regulation and social norms, and opposition to state-by-state AI laws; an editor may want to confirm those. Nvidia's own write-up of his January 2026 Davos remarks (https://blogs.nvidia.com/blog/davos-wef-blackrock-ceo-larry-fink-jensen-huang/, loaded) has further job-creation quotes. He has an obvious commercial interest in each of these positions.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Marc Andreessen

Cofounder and General Partner, Andreessen Horowitz (a16z)

Editorial estimate

Venture capitalist and leading techno-optimist. Predicts AI will bring an economic and jobs boom, calls fear of AI killing humanity a category error promoted by a cult, and argues the real danger is regulation that slows AI and entrenches incumbents.

“the idea that AI will decide to literally kill humanity is a profound category error”
Why AI Will Save the World · Andreessen Horowitz (a16z.com) · 6 Jun 2023

X · Economic impact

−85
Human complementarityHuman replacement
High confidence

Says AI will not cause mass unemployment and calls that fear the lump-of-labor fallacy. Expects productivity gains to raise demand, wages and employment, producing a jobs boom rather than job losses.

Evidence for this placement

'Why AI Will Save the World' (June 2023) argues AI 'may cause the most dramatic and sustained economic boom of all time', with record job and wage growth, and treats job-loss fears as the lump-of-labor fallacy [s1]. His manifesto says technological change increases the need for human work [s2]. Later reports show the same view: job-destruction fears are a 'fallacy' (October 2025) [s4] and job-loss narratives are 'all fake' (April 2026) [s5]. He has also said venture capital may outlast other, automated jobs [s6]. Emphatic: top complementarity band.

Y · Existential risk

−95
Manageable riskExistential threat
High confidence

Rejects the idea that AI could decide to kill humanity as a category error: AI is math and code controlled by people. Describes AI-risk advocacy as a cult and existential-risk thinking as a demoralizing bad idea.

Evidence for this placement

His June 2023 essay calls the idea that AI will decide to kill humanity 'a profound category error', calls the doomer position 'non-scientific' and says AI risk has developed into a cult [s1]. The October 2023 manifesto lists 'existential risk' among the names of a decades-long demoralization campaign against technology [s2]. On Lex Fridman's podcast in January 2025 he attributed other AI pioneers' 'doomerism' to their politics [s3]. Explicit, repeated dismissal of the risk as unfounded: far end of the scale.

Z · Regulatory intervention

−90
Minimal regulationStrong regulation
High confidence

Wants AI companies large and small free to build as fast as they can, with no regulatory barriers to open source. Regards AI regulation as a route to regulatory capture and says slowing AI costs lives.

Evidence for this placement

His 2023 essay proposes that big and startup AI companies be 'allowed to build AI as fast and aggressively as they can' and that open source face no regulatory barriers 'whatsoever', warning that AI-risk rules would create a government-protected cartel [s1]. The manifesto says 'any deceleration of AI will cost lives' [s2]. In January 2025 he described the rules sought by AI-risk advocates as 'ruinously destructive government legislation and regulation' [s3]. He treats regulation itself as the main danger: top minimal-intervention band.

“Startup AI companies should be allowed to build AI as fast and aggressively as they can.”
Why AI Will Save the World · Andreessen Horowitz (a16z.com) · 6 Jun 2023
Sources (6)
  1. [s1]
    Why AI Will Save the World
    Andreessen Horowitz (a16z.com) · 6 Jun 2023 · essay · primary source

    His long essay rebutting five claimed AI risks, including extinction and job loss, and setting out what he thinks policy should be.

    a16z.com
  2. [s2]
    The Techno-Optimist Manifesto
    Andreessen Horowitz (a16z.com) · 16 Oct 2023 · essay · primary source

    His manifesto on technology, markets and growth, with sections on AI, work and the ideas he regards as enemies of progress.

    a16z.com
  3. [s3]
    Transcript for Marc Andreessen: Trump, Power, Tech, AI, Immigration & Future of America | Lex Fridman Podcast #458
    Lex Fridman Podcast · 26 Jan 2025 · podcast · primary source

    Host-published transcript of a long interview published in January 2025; AI doomerism and regulation come up near the end.

    lexfridman.com
  4. [s4]
    Billionaire investor Marc Andreessen says AI destroying jobs and making everyone poor is a 'fallacy'—and even if that did happen, prices would drop
    Fortune · 8 Oct 2025 · news report · secondary source

    Report quoting his remarks about AI, jobs and prices on Stripe's Cheeky Pint podcast.

    fortune.com
  5. [s5]
    Marc Andreessen calls AI job loss fears fake, expects employment gains
    Cointelegraph · 6 Apr 2026 · news report · secondary source

    Report quoting a post he made on X predicting a jobs boom from AI-driven productivity.

    cointelegraph.com
  6. [s6]
    Marc Andreessen predicts one of the few jobs that may survive the rise of AI automation
    Fortune · 1 May 2025 · news report · secondary source

    Report on an a16z podcast remark that venture capital may be among the last fields people still work in once other jobs are automated.

    fortune.com
Editor’s notes and caveats

Views are stable from 2023 through 2026. The 2025 and 2026 jobs statements [s4], [s5], [s6] are news reports of podcasts and an X post that I did not load directly, so they are marked non-primary. [s6] reports him saying venture capital may be one of the last jobs left once others are automated, which sits awkwardly beside his no-job-loss forecasts; X is set at -85 rather than lower for that reason and is worth an editor's glance. His 2023 essay does accept a role for government in using AI defensively against misuse, so Z is not at the absolute pole. Reports of his funding of the Leading the Future super PAC and of a16z lobbying for federal preemption of state AI laws were seen only in search summaries and are not cited. Affiliation confirmed on his a16z team page.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Max Tegmark

Professor of physics, MIT; co-founder and president, Future of Life Institute

Editorial estimate

Sees the race to build AGI as a suicide race that humanity would probably lose control of, and thinks superintelligence would leave no jobs for people. Wants controllable tool AI under binding safety standards, with superintelligence prohibited until proven safe.

“People who claim there will still be jobs after superintelligence don't understand what superintelligence means.”
X post on jobs after superintelligence · X (formerly Twitter) · 5 Mar 2025

X · Economic impact

+72
Human complementarityHuman replacement
Medium confidence

Says AGI could do all human jobs and that none would remain after superintelligence, but argues this outcome is avoidable if society builds controllable tool AI instead.

Evidence for this placement

In a March 2025 X post he wrote that people who claim there will still be jobs after superintelligence do not understand what it means [s3]. CNN headlined a December 2025 interview with his prediction that superintelligence could make everybody economically obsolete [s7]. In 2022 he described AGI as machines that can do all our jobs [s8]. Full-substitution band. Medium, and placed at its lower edge, because the forecast is conditional: he argues such systems are not inevitable [s2] and that tool AI would bring shared prosperity [s1].

Y · Existential risk

+85
Manageable riskExistential threat
High confidence

Believes AGI is likely to arrive before anyone knows how to control it, making machine takeover the default outcome of the current race, though one that can be avoided by not building it.

Evidence for this placement

His 2024 essay says we are closer to building AGI than to controlling it, that takeover by a smarter-than-human species is the default unless a clever scheme appears, which he considers highly unlikely in time, and calls the race a suicide race [s1]. The Guardian reported in May 2025 that his own calculation gave a 90% probability that highly advanced AI would pose an existential threat [s6]. His 2023 Senate statement warns such systems could wipe out humanity [s2]. Catastrophe-by-default band.

Z · Regulatory intervention

+88
Minimal regulationStrong regulation
High confidence

Advocates licensing, independent audits, a federal AI regulator, compute registration and developer liability, plus a prohibition on building superintelligence until it can be shown to be safe and controllable.

Evidence for this placement

His 2023 Senate statement recommends audits and licensure before deployment, no licences for uncontrollable AI, registration of large compute purchases and training runs, a central federal authority and developer liability [s2]. His 2024 essay proposes binding FDA-style standards under which AGI stays paused until proven controllable, harmonised between the US and China [s1]. FLI, which he leads, published the 2023 pause letter [s5], and he signed the 2025 call to prohibit superintelligence development [s4]. Top band.

“this race is not an arms race but a suicide race”
The Hopium Wars: the AGI Entente Delusion · LessWrong · 13 Oct 2024
Sources (8)
  1. [s1]
    The Hopium Wars: the AGI Entente Delusion
    LessWrong · 13 Oct 2024 · essay · primary source

    Essay under his own name arguing against a US race with China to build AGI and proposing a 'tool AI' strategy with binding safety standards.

    lesswrong.com
  2. [s2]
    Written Statement of Dr. Max Tegmark, AI Insight Forum: Innovation
    U.S. Senate (office of Majority Leader Schumer) · 24 Oct 2023 · testimony · primary source

    Written statement submitted to the second bipartisan Senate AI Insight Forum, with five numbered policy recommendations for Congress.

    schumer.senate.gov
  3. [s3]
    X post on jobs after superintelligence
    X (formerly Twitter) · 5 Mar 2025 · social post · primary source

    Post sharing a documentary about human work in the age of AI. Timestamp is UTC.

    x.com
  4. [s4]
    Statement on Superintelligence
    Future of Life Institute · Oct 2025 · open letter · primary source

    Open statement calling for a prohibition on developing superintelligence until there is scientific consensus on safety and strong public buy-in. Tegmark is listed as a signatory.

    superintelligence-statement.org
  5. [s5]
    Pause Giant AI Experiments: An Open Letter
    Future of Life Institute · 22 Mar 2023 · open letter · primary source

    Open letter published by the institute he leads, calling for a six-month pause on training systems more powerful than GPT-4.

    futureoflife.org
  6. [s6]
    AI firms warned to calculate threat of super intelligence or risk it escaping human control
    The Guardian · 10 May 2025 · news report · secondary source

    News report by Dan Milmo on Tegmark's call for AI companies to calculate a 'Compton constant', the probability of losing control of superintelligence.

    theguardian.com
  7. [s7]
    Max Tegmark predicts superintelligence could make "everybody economically obsolete"
    CNN Business · 23 Dec 2025 · news report · secondary source

    Short video interview recorded at Web Summit Lisbon. Only the page headline and summary were read; the video itself was not transcribed.

    cnn.com
  8. [s8]
    Max Tegmark on how a 'put-up-or-shut-up' resolution led him to work on AI and algorithmic news selection
    80,000 Hours Podcast · 1 Jul 2022 · podcast · primary source

    Long-form interview with Rob Wiblin, with the podcast's own published transcript. Predates the preferred 2023 to 2026 window.

    80000hours.org
Editor’s notes and caveats

Y and Z are consistent across 2023 to 2025 primary sources. X is conditional rather than a flat forecast: he predicts total labor substitution if AGI or superintelligence is built, but campaigns to prevent that and expects controllable tool AI to complement people; an editor may prefer a value just under the top band. Could not verify: the April 2023 TIME essay ('The Don't Look Up Thinking That Could Doom Us With AI'), because time.com returned HTTP 406 and the Wayback Machine rate-limited requests, so it is not cited; and the wording of the CNN video [s7], where only the headline quotation was visible. The Guardian's 90% figure [s6] is reported speech, not a direct quotation. His name does not appear in the static HTML of the pause letter page [s5] (signatures load dynamically), so [s5] is cited as an FLI publication that he describes in [s2] rather than as a personally verified signature. His signature on [s4] appears in the signatory data embedded in the page (Professor of Physics, MIT; President, FLI; dated 2025-10-08). A third-party machine transcript of Lex Fridman Podcast #371 (April 2023) shows him making the same jobs argument, but it is too error-ridden to cite. Verified alternate quotations: from [s1], AGI approval should similarly require quantitative safety guarantees; from [s2], This future is simply not possible without robust regulation of the AI industry.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Mustafa Suleyman

CEO, Microsoft AI

Editorial estimate

DeepMind co-founder who now leads Microsoft AI. He predicts AI will soon handle most computer-based professional tasks, admits nobody can yet guarantee control of superintelligence, and argues for humanist AI kept subordinate to people, backed by regulation and independent audits.

“We simply don't know what might emerge from autonomous, constantly evolving and improving systems”
Towards Humanist Superintelligence · Microsoft AI · 6 Nov 2025

X · Economic impact

+55
Human complementarityHuman replacement
Medium confidence

Predicts human-level AI performance on most professional tasks and full automation of most computer-based white-collar tasks within about 18 months. He has called the tools fundamentally labour-replacing if left to the market.

Evidence for this placement

He told the Financial Times (Feb 2026), as reported by The Decoder, that most white-collar tasks 'will be fully automated by an AI within the next 12 to 18 months' [s5]. At Davos (Jan 2024), per Fortune, he called AI tools 'fundamentally labor replacing' if left to the market, while expecting augmentation for a couple of decades [s6]. His own essay says humanist AI should support human roles [s1], a design goal, not a forecast. Reported quotes only, so medium.

Y · Existential risk

+40
Manageable riskExistential threat
Medium confidence

Says no one has a reassuring answer for how to contain and align a self-improving superintelligence and calls that an urgent challenge. He rejects doom narratives and believes bounded, controllable systems can avoid the worst risks.

Evidence for this placement

He signed the 2023 extinction-risk statement [s4], yet that September told MIT Technology Review the existential-risk debate was a 'completely bonkers distraction' [s3]. His November 2025 essay asks how certain we are that we won't lose control and calls containment and alignment 'the urgent challenge facing humanity', while rejecting 'binaries of boom and doom' [s1]. On CNN (Sept 2026) he called recent agent incidents a 'watershed moment' [s2]. A shift toward concern: lower +30 to +69 band.

Z · Regulatory intervention

+45
Minimal regulationStrong regulation
High confidence

Unusually pro-regulation for an industry executive: he supports binding standards, independent audits of frontier labs, required audit logs and shutdown ability, and international safety accords. He has not backed a pause or a ban on superintelligence.

Evidence for this placement

MIT Technology Review (Sept 2023): 'I believe in the power of regulation', and self-improving AI might need to be a licensed activity [s3]. NPR (Dec 2024): he has long encouraged additional regulation [s7]. His 2025 essay welcomes accountability and oversight [s1]. On CNN (Sept 2026) he backed independent embedded evaluators, said industry should be required to keep tamper-proof audit logs, and wanted China in safety accords [s2]. The Statement on Superintelligence page quotes him but marks him as a non-signatory [s9]. Binding rules, not moratoria.

“The word regulation should not be a bad thing.”
Fareed Zakaria GPS: Government And Tech Leaders Reckon With A.I. Risks (rush transcript) · CNN · 20 Sep 2026
Sources (9)
  1. [s1]
    Towards Humanist Superintelligence
    Microsoft AI · 6 Nov 2025 · essay · primary source

    His bylined essay announcing the MAI Superintelligence Team and setting out a bounded, human-controlled vision of advanced AI.

    microsoft.ai
  2. [s2]
    Fareed Zakaria GPS: Government And Tech Leaders Reckon With A.I. Risks (rush transcript)
    CNN · 20 Sep 2026 · interview · primary source

    CNN's own transcript of his interview about AI-agent incidents, audits and regulation. Marked by CNN as a rush transcript.

    transcripts.cnn.com
  3. [s3]
    DeepMind's cofounder: Generative AI is just a phase. What's next is interactive AI.
    MIT Technology Review · 15 Sep 2023 · interview · primary source

    Published question-and-answer interview with Will Douglas Heaven, given while he was CEO of Inflection AI.

    technologyreview.com
  4. [s4]
    Statement on AI Extinction Risk
    Center for AI Safety · 30 May 2023 · open letter · primary source

    One-sentence statement that extinction risk from AI should be a global priority. He is listed as a signatory, then as CEO of Inflection AI.

    safe.ai
  5. [s5]
    Microsoft AI CEO: "Most" white-collar tasks will be automated in 18 months
    The Decoder · 13 Feb 2026 · news report · secondary source

    News report quoting his Financial Times interview; the FT original is paywalled and was not loaded.

    the-decoder.com
  6. [s6]
    DeepMind co-founder Mustafa Suleyman warns AI is a 'fundamentally labor replacing' tool over the long term
    Fortune · 17 Jan 2024 · news report · secondary source

    News report on his CNBC interview at the World Economic Forum in Davos.

    fortune.com
  7. [s7]
    From DeepMind to Google to Microsoft: AI pioneer Mustafa Suleyman (transcript)
    NPR, TED Radio Hour · 20 Dec 2024 · podcast · primary source

    NPR's transcript of an extended radio interview covering his career, AI companions and containment.

    npr.org
  8. [s8]
    U.S. should use Nvidia's powerful chips as a 'chokepoint' to force adoption of A.I. rules, DeepMind cofounder Mustafa Suleyman says
    Fortune · 4 Sep 2023 · news report · secondary source

    News report on a Financial Times interview in which he proposed tying access to advanced chips to safety commitments.

    fortune.com
  9. [s9]
    Statement on Superintelligence
    Future of Life Institute · Oct 2025 · open letter · primary source

    Public statement calling for a prohibition on developing superintelligence. Its page data shows a quotation attributed to him and flags him as not a signatory.

    superintelligence-statement.org
Editor’s notes and caveats

Y has moved. In September 2023 he dismissed existential-risk talk as a distraction; since late 2025 he writes about loss of control as an open and urgent problem. The placement follows the recent material, and medium confidence reflects the shift. X relies on reported quotes: the FT interview is paywalled and the Davos remarks were broadcast on CNBC. Headlines saying jobs will vanish go beyond his quoted words, which concern tasks. Note the tension between that forecast and Microsoft's stated aim of AI that serves people rather than replacing them. His book The Coming Wave was not consulted directly, so its containment proposals are not used as evidence. A CNBC appearance on 18 September 2026 reported by one outlet could not be confirmed; the CNN transcript covers the same argument. The CNN transcript is a rush transcript and may be revised. The quotation the FLI page attributes to him was not traced to its origin and is not used.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Nate Soares

President, Machine Intelligence Research Institute (MIRI)

Editorial estimate

Argues that a superintelligence grown with today's poorly understood methods would be indifferent to people and wipe them out as a side effect. Wants a global, verifiable halt to the superintelligence race, while chatbots, self-driving cars and narrow medical AI continue.

“the main target we should be shooting for is an indefinite pause on progress toward AGI”
Thoughts on the AI Safety Summit company policy requests and responses · Machine Intelligence Research Institute · 31 Oct 2023

X · Economic impact

+85
Human complementarityHuman replacement
Medium confidence

Accepts that past technology created more jobs than it destroyed, but says AI that outperforms people at everything would leave no place for human labor in the economy.

Evidence for this placement

On The Tucker Carlson Show in September 2026 he said he does not dispute the standard economic view of past automation, then argued that comparative advantage does not guarantee a survivable wage and that AIs more efficient at everything would have no use for people [s5]. MIRI's 2025 position paper, which he co-authored, makes the same horses-and-automobiles argument [s4]. That is the full-substitution band. Medium confidence: one explicit first-person statement, taken from an unofficial transcript, plus a co-authored paper.

Y · Existential risk

+95
Manageable riskExistential threat
High confidence

Believes superintelligence built with current methods would most likely kill everyone, through indifference rather than malice, and that AI lab leaders' own risk estimates are too low.

Evidence for this placement

The 2025 book he co-wrote is titled 'If Anyone Builds It, Everyone Dies' [s8]. In a 2025 Carnegie interview he said an indifferent superintelligence repurposing the world means humanity dies as a side effect, and called lab leaders' 25 to 30 percent risk figures low [s2]. In 2023 he wrote that any technical work moving closer to smarter-than-human AI carries unacceptable risk [s1]. MIRI's position paper says the expected outcome is extinction [s4]. Catastrophe by default: top band.

Z · Regulatory intervention

+95
Minimal regulationStrong regulation
High confidence

Wants an international, monitored and enforceable agreement to stop the race to superintelligence, including tracking of advanced chips, while letting chatbots, self-driving cars and narrow AI continue.

Evidence for this placement

In 2023 he wrote that the main target should be an indefinite pause on progress toward AGI [s1] and told AI developers to stop [s6]. He signed the October 2025 statement calling for a prohibition on developing superintelligence [s3]. In a 2025 Carnegie interview he backed an international treaty with chip monitoring, while saying chatbots and narrow medical AI need not be given up [s2]. In September 2026 he said his line is to shut the companies down, globally [s7]. Top band.

“humanity dies as a side effect”
Will AI Kill Us All? Nate Soares on His Controversial Bestseller · Carnegie Endowment for International Peace (The World Unpacked podcast) · 25 Sep 2025
Sources (8)
  1. [s1]
    Thoughts on the AI Safety Summit company policy requests and responses
    Machine Intelligence Research Institute · 31 Oct 2023 · essay · primary source

    Signed blog post assessing the UK AI Safety Summit's policy asks and seven AI companies' responses, written on the eve of the summit.

    intelligence.org
  2. [s2]
    Will AI Kill Us All? Nate Soares on His Controversial Bestseller
    Carnegie Endowment for International Peace (The World Unpacked podcast) · 25 Sep 2025 · podcast · primary source

    Interview with host Jon Bateman shortly after the book's release. Carnegie's own transcript, labelled AI-generated and possibly containing errors.

    carnegieendowment.org
  3. [s3]
    Statement on Superintelligence
    Future of Life Institute · Oct 2025 · open letter · primary source

    Open statement calling for a prohibition on developing superintelligence until there is scientific consensus on safety and strong public buy-in. Soares is listed as a signatory.

    superintelligence-statement.org
  4. [s4]
    The Problem
    LessWrong (crosspost of MIRI's position paper) · 5 Aug 2025 · essay · primary source

    MIRI's statement of its position, with Soares (as So8res), Yudkowsky and four colleagues on the byline. First posted on intelligence.org in February 2025.

    lesswrong.com
  5. [s5]
    Transcript: Nate Soares Interview on Tucker Carlson Show
    The Singju Post (unofficial transcript) · 11 Sep 2026 · interview · secondary source

    Third-party transcript of a nearly two-hour interview on The Tucker Carlson Show; the site gives September 11, 2026 as the premiere date.

    singjupost.com
  6. [s6]
    Apocalypse insurance, and the hardline libertarian take on AI risk
    LessWrong · 28 Nov 2023 · essay · primary source

    Essay under his So8res account arguing that even on libertarian principles AI developers are imposing uncompensated risk on everyone and should stop.

    lesswrong.com
  7. [s7]
    Breaking Points Interview: w/ AI Researcher Nate Soares on Slowing AI Development (Transcript)
    The Singju Post (unofficial transcript) · 14 Sep 2026 · interview · secondary source

    Third-party transcript of an interview with Krystal Ball and Saagar Enjeti; the site gives September 14, 2026 as the premiere date.

    singjupost.com
  8. [s8]
    Eliezer and I wrote a book: If Anyone Builds It, Everyone Dies
    LessWrong · 14 May 2025 · essay · primary source

    His own announcement, under his So8res account, of the book co-written with Eliezer Yudkowsky and published in September 2025.

    lesswrong.com
Editor’s notes and caveats

Y and Z are well documented in his own writing. X rests mainly on [s5], an unofficial transcript; the passage is explicit and matches the co-authored MIRI paper, but an editor should check it against the video (around 1:22 to 1:26). Verified alternate quotation for X from [s5], as transcribed: There'll be no place for us in the economy. Both 2026 Singju Post transcripts refer to recent AI incidents that I did not try to verify; no coordinate depends on them. The [s2] transcript is AI-generated by Carnegie, so the Y quotation should be checked against audio. The byline 'So8res' on [s4], [s6] and [s8] is his LessWrong account; [s8] is written in the first person as the book's co-author. The book's companion site ifanyonebuildsit.com sits behind a bot-check page and was not read, and the book text itself was not consulted. His signature on [s3] appears in the signatory data embedded in the page (title 'President', Machine Intelligence Research Institute, dated 2025-10-22). Title 'President' also per intelligence.org/team on 2026-10-13.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Nick Bostrom

Philosopher, Macrostrategy Research Initiative; formerly founding director, Future of Humanity Institute, University of Oxford

Editorial estimate

Expects superintelligence fairly soon and thinks it could go badly wrong, but argues the greater mistake would be never building it. Foresees a post-work world of abundance, and now warns that clumsy pauses or heavy-handed regulation could backfire.

“you could say that the goal of AI is full unemployment”
My chat (+transcript) with Nick Bostrom on life in an AI utopia · Faster, Please! (James Pethokoukis, Substack) · 6 Jun 2024

X · Economic impact

+85
Human complementarityHuman replacement
High confidence

Expects that if AI succeeds it will do essentially every job better and cheaper, leaving a post-work world with only niche demand for things people prefer to be done by humans.

Evidence for this placement

In a June 2024 interview he said that with true AGI all areas of human economic contribution are affected, that the goal of AI is full unemployment, and that the result is a post-work condition with a few exceptions where consumers prefer a human [s1]. His own page for Deep Utopia (2024) describes an era in which human labor becomes obsolete [s6]. In September 2026 he described skilled trades as valuable only during a transitional era [s5]. Explicit and repeated: full-substitution band.

Y · Existential risk

+45
Manageable riskExistential threat
High confidence

Regards misaligned superintelligence as a real existential risk that could eliminate humanity, but not as the default outcome; thinks the risk can be reduced and is worth running carefully.

Evidence for this placement

In 2024 he said the transition carries significant risks, including existential ones, yet that never developing superintelligence would be a catastrophe [s1]. His 2026 paper says building it is not like Russian roulette but like risky surgery for an otherwise fatal condition [s2]. In September 2026 he listed alignment failure leading to human elimination as the first class of disaster, while saying public concern is approaching its optimal level [s5]. Serious but solvable: lower half of the +30 to +69 band.

Z · Regulatory intervention

+10
Minimal regulationStrong regulation
Medium confidence

Wants somewhat more government oversight than exists, such as safety standards and monitoring, and room to pause briefly at critical moments, but is wary of moratoria, nationalisation and regulatory overshoot.

Evidence for this placement

In 2024 he said his optimal level of oversight and regulation is more than we currently have, but that he worries about overshoot [s1], and that he is skeptical of proposals risking a permanent ban [s4]. His 2025 paper has governments setting safety standards and monitoring while resisting nationalisation [s3]. His 2026 paper says poorly implemented pauses could do more harm than good and declines to endorse specific policy [s2]. Targeted rules plus opposition to heavier mandates is the mixed band; medium because the stance has shifted and is hedged.

“poorly implemented pauses could do more harm than good”
Optimal Timing for Superintelligence: Mundane Considerations for Existing People · nickbostrom.com (working paper, v1.0) · 2026
Sources (6)
  1. [s1]
    My chat (+transcript) with Nick Bostrom on life in an AI utopia
    Faster, Please! (James Pethokoukis, Substack) · 6 Jun 2024 · podcast · primary source

    Podcast interview about Deep Utopia with the host's own published transcript, covering automation, risk and regulation.

    fasterplease.substack.com
  2. [s2]
    Optimal Timing for Superintelligence: Mundane Considerations for Existing People
    nickbostrom.com (working paper, v1.0) · 2026 · paper · primary source

    Working paper modelling when superintelligence should be developed from the standpoint of people alive today, responding to calls for a pause or ban.

    nickbostrom.com
  3. [s3]
    Open Global Investment as a Governance Model for AGI
    nickbostrom.com (working paper, v1.15) · Oct 2025 · paper · primary source

    Working paper proposing that AGI be developed by widely held corporations inside a government-defined regulatory framework. First version July 2025.

    nickbostrom.com
  4. [s4]
    Nick Bostrom Made the World Fear AI. Now He Asks: What if It Fixes Everything?
    WIRED · 2 May 2024 · interview · primary source

    Question-and-answer interview with Will Knight about Deep Utopia, lightly edited for length and clarity.

    wired.com
  5. [s5]
    Transcript: Prof. Nick Bostrom on When AI Escapes Our Control – Peter McCormack Show
    The Singju Post (unofficial transcript) · 24 Sep 2026 · interview · secondary source

    Third-party transcript of a podcast interview; the site gives September 24, 2026 as the premiere date.

    singjupost.com
  6. [s6]
    Deep Utopia: Life and Meaning in a Solved World
    nickbostrom.com (book page; Ideapress, 2024) · 2024 · book · primary source

    His own website's description of the 2024 book. The book text itself was not consulted.

    nickbostrom.com
Editor’s notes and caveats

Clear shift in emphasis since Superintelligence (2014): he still affirms existential risk but since 2023 to 2024 has stressed the cost of never building superintelligence and the danger of over-regulation, saying in [s1] and [s4] that a permanent ban on AI, though still unlikely, looks more probable than it did two years earlier. Z is the judgement call: he supports more oversight than today and a short pause at the critical stage, yet argues against government moratoria, international prohibition and nationalisation, and [s2] says its analysis does not imply support for any particular policy; anything from about 0 to +25 is defensible. X is conditional on AI succeeding, which he treats as the main line. Y: he declines to give a probability. Not quoted: a sentence in [s4] reads as if it has a transcription slip ('wouldn't be an immense tragedy'), and the September 2025 London Standard interview in which he reportedly repeated 'The goal is full unemployment' returned HTTP 403, so it is not cited. [s5] is an unofficial transcript and should be checked against the recording. No day or month of publication is given on the 2026 paper [s2]. Verified alternate quotations: from [s2], Developing superintelligence is not like playing Russian roulette; from [s1], My probably optimal level of government oversight and regulation would be more than we currently have.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Sam Altman

Co-founder and CEO, OpenAI

Editorial estimate

Altman expects superintelligence within a few years and massive prosperity to follow. He predicts painful job churn but new kinds of work, treats loss of control as a real but solvable danger, and now favours lighter rules than the licensing he proposed in 2023.

“I have no fear that we’ll run out of things to do”
The Intelligence Age · Sam Altman (personal site) · 23 Sep 2024

X · Economic impact

−20
Human complementarityHuman replacement
Medium confidence

Expects whole classes of jobs to disappear but people to find new and different work, and says a jobs apocalypse is unlikely. He still expects a new social contract and possible intervention on capital versus labor.

Evidence for this placement

In 2024 he wrote he had no fear of running out of things to do [s1]. In 2025 he said whole classes of jobs will go, new ones will appear, and a new social contract will emerge gradually [s2][s3]. Reuters-sourced reporting from May 2026 has him ruling out a jobs apocalypse and admitting earlier fears were off [s11]. Mixed band, leaning complementarity; his 2021 essay predicted labor costs falling toward zero [s14], so the forecast has softened.

Y · Existential risk

+30
Manageable riskExistential threat
Medium confidence

Treats misalignment and loss of control as real dangers that must be solved first, but is confident they can be and expects the transition to feel manageable.

Evidence for this placement

Signed the 2023 extinction-risk statement [s6]. His 2025 plan starts with solving the alignment problem, yet predicts the singularity will feel manageable [s3]. In April 2026 he wrote that fear of AI is justified [s4]; in September 2026 he agreed labs must pace the frontier [s10]; in October he reportedly said serious loss of control is a risk the world should not accept [s12]. A 2015 post called superhuman AI probably the greatest threat to humanity [s15]. Lower edge of the serious-but-solvable band.

Z · Regulatory intervention

+10
Minimal regulationStrong regulation
Medium confidence

Supports federal testing, auditing and transparency for frontier models but not regulators approving or blocking releases. He describes his own approach as lighter touch than Anthropic's.

Evidence for this placement

In 2023 testimony he urged licensing or registration for models above a capability threshold [s5]. By May 2025 he urged only sensible regulation that would not slow America down [s7]. In 2026 OpenAI proposed mandatory federal evaluations without power to block releases [s8]; he backed federal legislation [s9], endorsed pacing the frontier [s10] and floated an IAEA-style body [s13], yet told Politico he favours lighter-touch rules and accepting some harms [s16]. Targeted rules, heavier mandates resisted, record mixed: the -29 to +29 band.

“It is important that the democratic process remains more powerful than companies.”
Untitled blog post (title shown as “-”) · Sam Altman's blog · 10 Apr 2026
Sources (16)
  1. [s1]
    The Intelligence Age
    Sam Altman (personal site) · 23 Sep 2024 · essay · primary source

    Personal essay predicting superintelligence within a few thousand days and massive prosperity, with a passage on labor-market change.

    ia.samaltman.com
  2. [s2]
    Three Observations
    Sam Altman's blog · 9 Feb 2025 · essay · primary source

    Blog essay on the economics of AGI, virtual co-workers, and the risk that the balance between capital and labor gets disrupted.

    blog.samaltman.com
  3. [s3]
    The Gentle Singularity
    Sam Altman's blog · 10 Jun 2025 · essay · primary source

    Blog essay arguing the takeoff has started; discusses jobs going away, a gradual new social contract, and solving alignment first.

    blog.samaltman.com
  4. [s4]
    Untitled blog post (title shown as “-”)
    Sam Altman's blog · 10 Apr 2026 · essay · primary source

    Personal statement of beliefs written after an attack on his home: justified fear of AI, safety, democratised AI, democratic control over companies.

    blog.samaltman.com
  5. [s5]
    Written Testimony of Sam Altman, hearing on “Oversight of A.I.: Rules for Artificial Intelligence”
    U.S. Senate Committee on the Judiciary, Subcommittee on Privacy, Technology, and the Law · 16 May 2023 · testimony · primary source

    Prepared Senate testimony calling regulation essential and proposing licensing or registration for models above a capability threshold.

    judiciary.senate.gov
  6. [s6]
    Statement on AI Extinction Risk
    Center for AI Safety · May 2023 · open letter · primary source

    One-sentence open statement that mitigating extinction risk from AI should be a global priority; Altman is listed among the signatories.

    safe.ai
  7. [s7]
    Sam Altman urges lawmakers against regulations that could ‘slow down’ U.S. in AI race against China
    Fortune · 8 May 2025 · news report · secondary source

    Report on his Senate Commerce Committee appearance, contrasting his call for regulation that does not slow the US down with his 2023 testimony.

    fortune.com
  8. [s8]
    OpenAI responds to White House executive order on AI governance
    CSO Online · 4 Jun 2026 · news report · secondary source

    Report quoting OpenAI's federal framework blueprint: mandatory pre-release government evaluations, but no authority to approve or block deployments.

    csoonline.com
  9. [s9]
    Sam Altman Voices Support for Federal AI Guardrails During DC Visit
    PYMNTS · 29 Jul 2026 · news report · secondary source

    Summary of Bloomberg and Politico reporting on his remarks to reporters after meeting lawmakers in Washington.

    pymnts.com
  10. [s10]
    Tech leaders and governments split over ‘AI doom’ fears
    Reuters, via TimesLIVE · 21 Sep 2026 · news report · secondary source

    Wire roundup quoting his X post agreeing with Dario Amodei that the industry needs to pace the frontier.

    timeslive.co.za
  11. [s11]
    OpenAI’s Sam Altman: AI will not trigger a global ‘jobs apocalypse’
    Capacity · 26 May 2026 · news report · secondary source

    Report relaying Reuters coverage of his remarks to a Commonwealth Bank of Australia conference in Sydney.

    capacityglobal.com
  12. [s12]
    Sam Altman tells Politico the world “should accept some harm from AI”
    The Next Web · 5 Oct 2026 · news report · secondary source

    Report on his interview for Politico's Decoded newsletter, including his limits on which AI risks the world should accept.

    thenextweb.com
  13. [s13]
    India’s leadership, AI regulation, super intelligence and the future: What Sam Altman said at AI Summit
    Kashmir Reader (agency copy) · 20 Feb 2026 · news report · secondary source

    Agency report on his India AI Impact Summit keynote in New Delhi, covering jobs, superintelligence and an IAEA-style body.

    kashmirreader.com
  14. [s14]
    Moore's Law for Everything
    Sam Altman (personal site) · 16 Mar 2021 · essay · primary source

    Earlier essay predicting AI will do more and more human work and proposing taxes on capital and land to share the wealth.

    moores.samaltman.com
  15. [s15]
    Machine intelligence, part 1
    Sam Altman's blog · 25 Feb 2015 · essay · primary source

    Early blog post, written before OpenAI was founded, on why superhuman machine intelligence should be feared.

    blog.samaltman.com
  16. [s16]
    'Freaking Insane': Sam Altman Sparks Horror by Saying We Should 'Accept Some Bad Things Happening' for Sake of AI
    Common Dreams · 5 Oct 2026 · news report · secondary source

    Critical report that quotes his Politico Decoded interview at length, including his reference to lighter-touch regulation.

    commondreams.org
Editor’s notes and caveats

Z has moved twice: licensing in 2023, a light-touch stance in 2025, then partial movement back in 2026 (mandatory evaluations, pacing, an IAEA-style body) while still calling his approach lighter touch than Anthropic's. Y is borderline between the modest and serious bands; +25 would also be defensible. Not verified at primary level: the Politico Decoded interview, his 12 September 2026 X post, the Bloomberg and Reuters originals, OpenAI's June 2026 blueprint and April 2026 paper 'Industrial Policy for the Intelligence Age' (openai.com, politico.com and axios.com refused automated requests), and the oral Q&A of the May 2025 Senate hearing; all of these rest on reporting. The OpenAI papers are company documents, not first-person statements. The 2023 testimony was read from the Senate PDF. The date of the untitled blog post comes from the page's embedded timestamp (2026-04-10 UTC).

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Sayash Kapoor

Incoming Assistant Professor, UC Berkeley School of Information (from July 2027); recent computer science Ph.D., Princeton University Center for Information Technology Policy

Editorial estimate

Co-author of AI Snake Oil and AI as Normal Technology. Argues AI's effects arrive slowly through adoption, so jobs get redefined rather than wiped out; that rogue-AI scenarios are a security and control problem, not doom; and that policy should build resilience and accountability.

“like previous waves of automation, we think the definitions of those jobs will change”
Computer scientist on why he believes mass layoffs due to AI is a 'convenient excuse' · NPR Morning Edition · 30 Jun 2026

X · Economic impact

−60
Human complementarityHuman replacement
High confidence

Expects AI to redefine jobs rather than eliminate them, with new products and markets outweighing labor substitution. Dismisses most AI-layoff stories as a corporate excuse, but worries about fewer entry-level openings.

Evidence for this placement

On NPR in June 2026 [s1] he called AI a 'convenient excuse' for layoffs, said job definitions will change as in earlier automation waves, and said he would bet on new markets over labor substitution, while calling weaker junior hiring worth taking seriously and predicting dramatic changes in many fields. Co-authored essays say automation will not make human labor superfluous [s3] and reject the mass-layoffs narrative [s9]. Mostly augmentation with real disruption: -69 to -30 band, placed just below Narayanan because of his explicit entry-level concern.

Y · Existential risk

−60
Manageable riskExistential threat
High confidence

Rejects the idea that loss of control leads to endless, self-amplifying catastrophe. Sees rogue-agent incidents as security failures that better control engineering and policy can fix; catastrophe is not imminent, but risks rise without investment.

Evidence for this placement

As first author of the September 2026 essay on loss-of-control incidents [s2] he writes that the authors strongly disagree with the superintelligence view of endless risk, that catastrophic cyber-risk is multiple qualitative changes away, and that they reject alarmism but agree safety is not on track. The 2025 essay calls catastrophic misalignment the most speculative risk [s3]; the October 2026 essay calls x-risk warnings sincere but wrong [s5]. In 2023 he said most researchers are far from thinking x-risk is imminent [s6]. Unlikely and manageable: -69 to -30 band.

Z · Regulatory intervention

−10
Minimal regulationStrong regulation
Medium confidence

Supports liability for AI companies, mandatory transparency and incident reporting, audits, safe harbors for researchers and DNA-synthesis screening. Opposes licensing, nonproliferation and limits on releasing models as costly, hard to enforce and power-concentrating.

Evidence for this placement

First author of a 2023 post against licensing [s7] and a May 2026 essay against extraordinary interventions such as release restrictions, which calls chip export controls a modest step [s4]. First author of the September 2026 essay saying AI companies should be liable for their agents and that regulating developers matters more than the authors had argued [s2]. Signed a June 2026 open letter for mandatory nucleic acid synthesis screening [s8]. Targeted rules and liability without heavier mandates: mixed, -29 to +29 band, up from about -30 before 2026.

“We think AI companies should be liable for what their agents do”
The AI-as-Normal-Technology view of loss-of-control incidents · AI as Normal Technology (normaltech.ai) · 14 Sep 2026
Sources (10)
  1. [s1]
    Computer scientist on why he believes mass layoffs due to AI is a 'convenient excuse'
    NPR Morning Edition · 30 Jun 2026 · interview · primary source

    Official NPR transcript of a solo radio interview with Steve Inskeep about AI and layoffs.

    npr.org
  2. [s2]
    The AI-as-Normal-Technology view of loss-of-control incidents
    AI as Normal Technology (normaltech.ai) · 14 Sep 2026 · essay · primary source

    13,000-word essay (Kapoor first author, with Narayanan) responding to 2026 rogue-agent incidents; includes a section on how their views changed.

    normaltech.ai
  3. [s3]
    AI as Normal Technology
    Knight First Amendment Institute at Columbia University · 15 Apr 2025 · essay · primary source

    Long essay co-authored with Arvind Narayanan setting out their worldview on AI's pace, risks and policy.

    knightcolumbia.org
  4. [s4]
    Do AI Risks Require Extraordinary Government Intervention?
    AI as Normal Technology (normaltech.ai) · 21 May 2026 · essay · primary source

    Newsletter essay (Kapoor first author, with Narayanan) replying to Derek Thompson on whether AI risks justify restricting releases.

    normaltech.ai
  5. [s5]
    A big-tent or small-tent AI safety movement?
    AI as Normal Technology (normaltech.ai) · 1 Oct 2026 · essay · primary source

    Newsletter essay co-authored with Narayanan, written after a wave of public existential-risk warnings in September 2026.

    normaltech.ai
  6. [s6]
    The Princeton researchers calling out 'AI snake oil'
    Semafor · 15 Sep 2023 · interview · primary source

    Published Q&A with Narayanan and Kapoor; answers are attributed to each speaker individually.

    semafor.com
  7. [s7]
    Licensing is neither feasible nor effective for addressing AI risks
    AI Snake Oil newsletter (now normaltech.ai) · 10 Jun 2023 · essay · primary source

    Newsletter post (Kapoor first author, with Narayanan) arguing against licensing and nonproliferation proposals for AI.

    normaltech.ai
  8. [s8]
    An Open Letter in Support of Mandatory Nucleic Acid Synthesis Screening and Recordkeeping
    screendna.org · Jun 2026 · open letter · primary source

    Open letter to legislators from researchers and AI and biotech figures; Kapoor appears in the signatory list.

    screendna.org
  9. [s9]
    Why AI hasn't replaced software engineers, and won't
    AI as Normal Technology (normaltech.ai) · 11 Jun 2026 · essay · primary source

    Newsletter essay co-authored with Narayanan examining claimed AI-driven layoffs in software engineering.

    normaltech.ai
  10. [s10]
    Sayash Kapoor Joins UC Berkeley with Focus on Artificial Intelligence
    UC Berkeley School of Information · 28 Jul 2026 · news report · secondary source

    Institutional announcement of his appointment; used only to confirm current affiliation, not for any coordinate.

    ischool.berkeley.edu
Editor’s notes and caveats

Affiliation changed in 2026: the Berkeley announcement [s10] and his own Princeton-hosted bio page (https://www.cs.princeton.edu/~sayashk/bio.html, loaded) say he is an incoming assistant professor at the UC Berkeley School of Information from July 2027, finished his Princeton Ph.D. requirements in August 2026, was previously a Mozilla senior fellow, and is writing a book with Narayanan in the interim. Most evidence is co-authored with Narayanan; solo evidence is the NPR interview [s1], his Semafor answers [s6] and the open-letter signature [s8]. Y and Z are set equal to Narayanan's because nothing found separates them; the 5-point X difference rests on one interview and an editor may reasonably equalize it. Z confidence is medium because the authors themselves describe a 2026 update toward regulating developers. Not checked for lack of transcripts: September 2026 WHYY, NBC News and The Hindu appearances on existential risk, the Lawfare and Carnegie India podcasts (summary pages only), and Hard Fork Live.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Yann LeCun

Executive Chairman, AMI Labs (Advanced Machine Intelligence); Professor, New York University

Editorial estimate

Turing Award winner who left Meta to found AMI Labs. Argues AI will amplify human intelligence rather than replace workers, calls extinction scenarios preposterous, champions open-source models, and says rules belong on products, not on research and development.

“There is no reason to believe that just because AI systems are intelligent they will want to dominate us.”
How Not to Be Stupid About AI, With Yann LeCun · WIRED · 22 Dec 2023

X · Economic impact

−75
Human complementarityHuman replacement
High confidence

Expects AI to be another set of tools that makes people more productive. Professions will shift gradually, as in earlier technological revolutions, without mass or permanent unemployment.

Evidence for this placement

He told the BBC in June 2023 that AI is 'not going to put a lot of people out of work permanently' [s3]. On Lex Fridman's podcast (March 2024) he said economists he consults expect no mass unemployment, only a gradual shift of professions, and that AI will amplify human intelligence [s1]. In May 2026 he called job-apocalypse warnings 'extremely destructive' and said AI is 'just another set of tools' [s6]. Emphatic rejection of mass-unemployment forecasts: top complementarity band, tempered because he defers to economists.

Y · Existential risk

−90
Manageable riskExistential threat
High confidence

Dismisses extinction and takeover scenarios as preposterous. Says intelligence does not imply a drive to dominate, superintelligence will arrive gradually, and systems will be built with guardrails and stay under human control.

Evidence for this placement

On Lex Fridman (March 2024) he said doomer scenarios rest on assumptions that are 'mostly false' and called the idea that smarter AI would eliminate us 'preposterous' [s1]. He told WIRED (December 2023) there is no reason intelligent systems would want to dominate us [s2], and the BBC that takeover fears are 'preposterously ridiculous' [s3]. Fortune (October 2026) reports he has 'zero concerns' about recent rogue-agent incidents and is not worried 'at all' about extinction [s5]. Consistent outright dismissal: top band.

Z · Regulatory intervention

−60
Minimal regulationStrong regulation
High confidence

Accepts safety rules for specific products and applications, where he says regulation already exists, but opposes regulating AI research and development or open-source models. Sees calls to regulate frontier models as regulatory capture.

Evidence for this placement

He told WIRED (December 2023) that product-safety regulation 'makes sense' but regulating AI research and development does not [s2], and the BBC that each application needs its own rules [s3]. He publicly opposed California's SB 1047 in 2024 [s7]. His 2023 Senate testimony, given as Meta's chief scientist, welcomed transparency and red-teaming standards [s4]. Fortune (October 2026) reports he thinks AI needs no new regulation and calls the push for it 'regulatory capture' [s5]. Light-touch band, toward its strong end.

“AI basically will amplify human intelligence.”
Transcript for Yann Lecun: Meta AI, Open Source, Limits of LLMs, AGI & the Future of AI | Lex Fridman Podcast #416 · Lex Fridman Podcast · 8 Mar 2024
Sources (8)
  1. [s1]
    Transcript for Yann Lecun: Meta AI, Open Source, Limits of LLMs, AGI & the Future of AI | Lex Fridman Podcast #416
    Lex Fridman Podcast · 8 Mar 2024 · podcast · primary source

    Host-published transcript of a long interview recorded while LeCun was Meta's chief AI scientist; covers AI doomers, open source and jobs.

    lexfridman.com
  2. [s2]
    How Not to Be Stupid About AI, With Yann LeCun
    WIRED · 22 Dec 2023 · interview · primary source

    Question-and-answer interview with Steven Levy covering superintelligence, open source, the EU AI Act and whether research should be regulated.

    wired.com
  3. [s3]
    Meta scientist Yann LeCun says AI won't destroy jobs forever
    BBC News · 15 Jun 2023 · news report · secondary source

    BBC report with direct quotes from a Meta press event in Paris, held ahead of a European Parliament vote on the AI Act.

    bbc.com
  4. [s4]
    Testimony of Yann LeCun, Chief AI Scientist, Meta
    US Senate Select Committee on Intelligence · 19 Sep 2023 · testimony · primary source

    Written statement for a hearing on the national security implications of AI, given in his capacity as Meta's chief AI scientist.

    intelligence.senate.gov
  5. [s5]
    AI 'godfather' Yann LeCun has 'zero concerns' about human extinction, says Anthropic CEO Dario Amodei is 'deluded'
    Fortune · 1 Oct 2026 · news report · secondary source

    Feature by Emily Forlini based on an in-person interview in New York; mixes direct quotes with the reporter's paraphrase of his views on regulation.

    fortune.com
  6. [s6]
    An AI 'godfather' says CEOs hyping job loss are 'extremely destructive'—and your kids are paying the price
    Fortune · 5 May 2026 · news report · secondary source

    Report quoting LeCun's remarks to Axios about AI job-loss warnings and their effect on students.

    fortune.com
  7. [s7]
    AI safety showdown: Yann LeCun slams California's SB 1047 as Geoffrey Hinton backs new regulations
    VentureBeat · 11 Sep 2024 · news report · secondary source

    Report on LeCun's public posts criticizing supporters of California's frontier-model safety bill.

    venturebeat.com
  8. [s8]
    Yann LeCun's new venture is a contrarian bet against large language models
    MIT Technology Review · 22 Jan 2026 · interview · primary source

    Edited question-and-answer interview about leaving Meta and founding AMI Labs; he says he will be executive chairman and keep his NYU post.

    technologyreview.com
Editor’s notes and caveats

Affiliation confirmed on his own home page (yann.lecun.com lists Executive Chairman of AMI Labs and his NYU professorship) and in [s8]. Z deserves a second look: the 2023 Senate statement was given for Meta and is friendlier to regulation than his own interviews, while Fortune's October 2026 line that he sees no need for new AI regulation is the reporter's paraphrase; if that hardening is confirmed he would sit nearer -70. Part of the jobs evidence in [s1] is him relaying what economists tell him. Not verified this session: the Munk Debate (no accessible transcript found), TIME's February 2024 interview and the original Axios interview behind [s6] (both blocked to curl), and his own posts on SB 1047 (seen only through [s7]). [s5] and [s6] describe 2026 events as reported by Fortune; I did not corroborate those events independently.

Last reviewed 13 Oct 2026 · Dataset v1

Thinker placements are evidence-based editorial interpretations of public statements. They are not self-reported, and none of these people took the questionnaire.

Yoshua Bengio

Full Professor of Computer Science, Université de Montréal; Co-President and Scientific Director, LawZero; Founder and Scientific Advisor, Mila

Editorial estimate

Turing Award winner who turned to AI safety in 2023. He argues that agentic AI could slip out of human control, wants frontier AI licensed and governed by international agreement, and founded the nonprofit LawZero to build safe-by-design systems.

“Frontier AI should be licensed like other critical technologies”
Professor Yoshua Bengio's speech at the UN Security Council · yoshuabengio.org · 23 Sep 2026

X · Economic impact

+70
Human complementarityHuman replacement
Medium confidence

Expects AI eventually to do most jobs people do today: keyboard-based cognitive work first, physical work later as robotics catches up. He says retraining will not save those displaced and that wealth will have to be redistributed.

Evidence for this placement

On Diary of a CEO (Dec 2025) he said it is 'a matter of time' before AI can do most of the jobs people do, and that the relative safety of physical work is 'only a temporary thing' [s5]; Fortune reported the same remarks [s8]. In a February 2026 Moneycontrol interview, as reported by Storyboard18, he said more and more jobs will be automated and called for redistribution [s9]. One unofficial transcript and two reports, hence medium.

Y · Existential risk

+65
Manageable riskExistential threat
High confidence

Treats loss of human control over advanced AI as a real and possibly near danger that could end in catastrophe, and says even a small chance is unacceptable. He believes AI that is safe by design can be built.

Evidence for this placement

Signed the 2023 extinction-risk statement [s7]. His July 2024 essay argues the case for not worrying fails, given stakes up to human extinction [s3]. Launching LawZero (June 2025) he wrote that the risk of losing control is 'all too real' and compared the situation to Russian roulette [s2]. At the UN Security Council (Sept 2026) he named loss of control as one of three global threats, yet said he is confident controllable AI can be created [s1]. Upper +30 to +69 band.

Z · Regulatory intervention

+85
Minimal regulationStrong regulation
High confidence

Advocates licensing of frontier AI, independent proof of safety before training and deployment, mandatory liability insurance, and verified international agreements. He also signed a call to prohibit superintelligence development until it is shown to be safe.

Evidence for this placement

Senate testimony (July 2023) urged licences, independent audits and international regulatory frameworks [s4]. His 2024 essay defends regulation and treaties against common objections [s3]. He co-signed the SB 1047 support letter [s10] and the 2025 Statement on Superintelligence, which calls for a prohibition [s6]. At the UN Security Council (Sept 2026) he said frontier AI should be licensed, liability insurance required, and that no company or country should build systems capable of large-scale harm [s1]. Treaties plus a prohibition call: top band.

“the risk of losing control is all too real”
Introducing LawZero · yoshuabengio.org · 3 Jun 2025
Sources (10)
  1. [s1]
    Professor Yoshua Bengio's speech at the UN Security Council
    yoshuabengio.org · 23 Sep 2026 · talk · primary source

    Text of his address to the UN Security Council on uncontrolled frontier AI, posted on his own blog.

    yoshuabengio.org
  2. [s2]
    Introducing LawZero
    yoshuabengio.org · 3 Jun 2025 · essay · primary source

    Blog post announcing his nonprofit AI safety lab and the non-agentic Scientist AI research plan.

    yoshuabengio.org
  3. [s3]
    Reasoning through arguments against taking AI safety seriously
    yoshuabengio.org · 9 Jul 2024 · essay · primary source

    Long essay answering common objections to AI safety, regulation and international treaties.

    yoshuabengio.org
  4. [s4]
    Oversight of A.I.: Principles for Regulation (Senate Hearing 118-108)
    U.S. Government Publishing Office · 25 Jul 2023 · testimony · primary source

    Official record of the Senate Judiciary subcommittee hearing at which he testified alongside Dario Amodei and Stuart Russell.

    govinfo.gov
  5. [s5]
    Creator of AI: We Have 2 Years Before Everything Changes! These Jobs Won't Exist in 24 Months! (transcript)
    The Diary Of A CEO with Steven Bartlett, transcript hosted by PodScripts · 18 Dec 2025 · podcast · primary source

    Long-form podcast interview. The page is an unofficial machine transcript with visible errors, so no quotation is taken from it.

    podscripts.co
  6. [s6]
    Statement on Superintelligence
    Future of Life Institute · Oct 2025 · open letter · primary source

    One-sentence public statement calling for a prohibition on developing superintelligence. The page's signatory data lists Bengio as a verified signer.

    superintelligence-statement.org
  7. [s7]
    Statement on AI Extinction Risk
    Center for AI Safety · 30 May 2023 · open letter · primary source

    One-sentence statement that extinction risk from AI should be a global priority. Bengio is the second listed signatory.

    safe.ai
  8. [s8]
    AI pioneer Yoshua Bengio says it's 'only a matter of time' before AI wipes out every job
    Fortune · 26 Mar 2026 · news report · secondary source

    News report on his Diary of a CEO remarks about jobs; a re-run of a story first published 19 December 2025.

    fortune.com
  9. [s9]
    'Governments are not doing anything': Yoshua Bengio warns of AI job losses
    Storyboard18 · 18 Feb 2026 · news report · secondary source

    Report of his Moneycontrol interview on the sidelines of the India AI Impact Summit in Delhi.

    storyboard18.com
  10. [s10]
    Exclusive: Renowned Experts Pen Support for California's Landmark AI Safety Bill
    TIME · 7 Aug 2024 · news report · secondary source

    News report quoting a letter from Bengio, Hinton, Lessig and Russell urging California leaders to pass SB 1047.

    time.com
Editor’s notes and caveats

X rests on one podcast interview (unofficial transcript) and one reported interview; no essay or testimony of his own on labour markets was found, so confidence is medium even though the statements are explicit. Fortune's headline ('every job') is stronger than his quoted words ('most of the jobs'). Y: he generally declines to give one probability and argues from unacceptable stakes; a figure of about 20 percent attributed to him elsewhere was not verified this session. The International AI Safety Report, which he chairs, was deliberately not used because it is a multi-author synthesis. His September and October 2026 posts describe recent AI-agent incidents that I did not verify independently; they are cited only for his stated positions. The Statement on Superintelligence date is given to the month (see the Hinton entry).