ClaimsPeople

Eliezer Yudkowsky

Eliezer Yudkowsky founded the Machine Intelligence Research Institute (MIRI) and LessWrong, co-authored the best-selling book 'If Anyone Builds It, Everyone Dies' (Little, Brown, 2025), and was named to the TIME100 AI 2023 list.

2,016 claims collected, 236 researched, 79 resolved. Report of 4 November 2026, method 0.1.0+c876030c.

Scores by kind of claim

Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.

  • Factual claims

    Statements about the present or the past, checked against the record: did they hold up?

    76 70 82

    High confidence, 71 factual claims resolved. From 16 documents, 2007 to 2026; 18 documents sampled from 23 candidates found.

    Not scored: 50 unresolvable, 8 checkable only against their own records.

  • Predictions

    Statements about events outside the person's control, checked once they fell due: did they come true?

    60 42 78

    Low confidence, 8 predictions resolved. From 4 documents, 2008 to 2024; 18 documents sampled from 23 candidates found.

    Not scored: 97 unresolvable. 1 more fall due later.

    Does not yet allow for how likely each prediction was when made.

All types combined: 74 (likely 68 to 80, high confidence), from 79 claims: 71 factual claims, 8 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.

Eliezer Yudkowsky's claims

236 claims, sorted by newest said. Choose a claim to read it in full.
VerdictKindSaidClaimTopic
Resolved true Prediction 2008 Yudkowsky doubted that a thousandfold increase in AI hardware would buy anything like a thousandfold increase in effective intelligence, while noting that such doubt is not reliable without an analytical calculation. Prediction · 2008 AI compute, hardware and data
Unresolvable Prediction 2008 Yudkowsky said he thinks it will prove easier, in the end, to build a 747 from scratch than to scale up an existing bird or graft on jet engines, an analogy he used for building Artificial Intelligence rather than taking another road into the challenge of smarter-than-human intelligence. Prediction · 2008 AI timelines and forecasting
Unresolvable Prediction 2008 With molecular nanotechnology, an AI could potentially rewrite the solar system unopposed. Prediction · 2008 Other global risks
Own records only Factual 2008 When he first began thinking about these matters, Yudkowsky expected the largest benefits from Artificial Intelligence yet felt no anxiety about it. Factual · 2008 Their own work or organisation
Unresolvable Prediction 2008 Human uploading and upgrading will very probably not be the ground on which humanity first confronts smarter-than-human intelligence. Prediction · 2008 AI timelines and forecasting
Mostly true Factual 2008 The bystander effect is usually explained as resulting from diffusion of responsibility and pluralistic ignorance. Factual · 2008 Science and research methods
Unresolvable Prediction 2008 An AI might make a huge jump in intelligence after reaching some threshold of criticality. Prediction · 2008 AI timelines and forecasting
Unresolvable Factual 2008 The people who first invented and spread communism were idealists acting with good intentions. Factual · 2008 Philosophy, ethics and society
Unresolvable Prediction 2008 An Artificial Intelligence could rewrite its code from scratch and change the underlying dynamics of optimization. Prediction · 2008 AI timelines and forecasting
Unresolvable Prediction 2008 Yudkowsky said he thinks it is possible for fallible humans to succeed at building Friendly AI, but only if intelligence ceases to be a sacred mystery to us. Prediction · 2008 AI safety and alignment
Unresolvable Prediction 2008 There will not necessarily be distant advance warning before AI is created, so it cannot be relied upon. Prediction · 2008 AI timelines and forecasting
Unresolvable Prediction 2008 In Yudkowsky's opinion at the time, which he said did not rest on a precise calculation using a precisely confirmed theory, sharp jumps in intelligence are possible and likely, and constitute the dominant probability. Prediction · 2008 AI timelines and forecasting
Mostly true Factual 2008 The human brain is not end-user-modifiable. Factual · 2008 Science and research methods
Mostly true Prediction 2008 Most AI researchers will not want to make unFriendly AIs. Prediction · 2008 AI safety and alignment
Mostly true Factual 2008 Generations of investigators tried and failed to understand stars, chemistry and biology, which acquired a reputation of being impossible for science. Factual · 2008 Science and research methods
Mixed Factual 2008 The Kahneman, Slovic and Tversky edited volume helped establish the heuristics and biases field and was written with outside academic readers in mind. Factual · 2008 Science and research methods
Unresolvable Factual 2007 Humans were created by evolution, an unintelligent process, whereas AI will be created by intelligent human processes. Factual · 2007 AI capabilities
Unresolvable Prediction 2007 An AI project will not succeed merely by implementing a five-layer model of intelligence like the human brain's. Prediction · 2007 AI capabilities
Unresolvable Prediction 2007 A deliberately designed AI is likely to begin as a set of subsystems in a primitive, undeveloped state already designed to form a functioning supersystem. Prediction · 2007 AI capabilities
Unresolvable Factual 2007 Yudkowsky, on first visualizing a triangular light bulb, saw a pyramidal bulb with smoothed edges and a bulb on the square base. Factual · 2007 Philosophy, ethics and society
Unresolvable Prediction 2007 Yudkowsky conjectured that matching imagery against large concept sets will be one of the most computationally intensive subprocesses in AI, perhaps the most expensive subprocess. Prediction · 2007 AI capabilities
Unresolvable Factual 2007 Yudkowsky estimated that the whole process of forming the mental image in his example took less than one second. Factual · 2007 Philosophy, ethics and society
Unresolvable Prediction 2007 Cautiously considering what a more foresightful and less accretive design for intelligence might look like, Yudkowsky suggested that in an AI the center of gravity of intelligence, the process humans observe through their internal narrative, would probably become the central supersystem of the mind, to an even greater degree than in humans. Prediction · 2007 AI capabilities
Mostly true Factual 2007 The process of visualizing a triangular light bulb is not really open to human introspection. Factual · 2007 Philosophy, ethics and society
Unresolvable Factual 2007 Yudkowsky cannot introspect on how the salient facet of a concept is selected during concept combination. Factual · 2007 Philosophy, ethics and society
Unresolvable Factual 2007 The mental image of the pyramidal light bulb carried semantic content, such as knowledge of its glass composition, imported from the "light bulb" concept. Factual · 2007 Philosophy, ethics and society
Resolved true Factual 2007 The Copycat program has a limited store of preprogrammed concepts and does not learn further concepts through experience. Factual · 2007 AI capabilities
Unresolvable Prediction 2007 Early-stage AIs will likely be less adept than humans at visualizing concepts only as far as needed, and so may need full visualizations where humans use abstract reasoning. Prediction · 2007 AI capabilities
Mixed Factual 2007 Smooth fitness landscapes are a necessary and fundamental outcome of evolution. Factual · 2007 Science and research methods
Unresolvable Factual 2007 When a person reads the letters spelling "light bulb", the phonemes for it flow through the auditory cortex. Factual · 2007 Science and research methods
Unresolvable Prediction 2007 The distinction between sharp, fragile human-written code and smooth biological computation is unlikely to be erased by programmer effort or new programming languages. Prediction · 2007 AI capabilities
Mostly true Factual 2007 Under GOFAI and formal logic, the classical view of concepts treats abstraction as information loss. Factual · 2007 AI capabilities
Unresolvable Prediction 2007 Software-level parallelism, poorly handled by present-day languages, will likely be one of the greatest challenges in AI development. Prediction · 2007 AI capabilities
Unresolvable Factual 2007 The mental image of a triangular light bulb would intuitively appear to result from imposing "triangular" on the "light bulb" concept. Factual · 2007 Philosophy, ethics and society
Unresolvable Factual 2007 The idea of 'pure' abstract reasoning has historically given rise to pathologies in AI research. Factual · 2007 AI capabilities
Mostly true Factual 2007 Three color channels are not enough to discriminate between all natural surface reflectances. Factual · 2007 Science and research methods
Back to the list

Prediction AI safety and alignment

“Most AI researchers will not want to make unFriendly AIs.”

Eliezer Yudkowsky, author of chapter 2008 Artificial Intelligence as a Positive and Negative Factor in Global Risk (MIRI)

Most AI researchers will not want to make unFriendly AIs.

Mostly true True in substance, wrong in a detail.

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

Read 'unFriendly AI' as AI that is harmful to, or misaligned with, human interests. True: surveys or other systematic evidence show that a clear majority of AI researchers do not intend or want to build harmful or misaligned AI, for example by preferring beneficial or safe outcomes. Mostly true: the majority holds, but the evidence covers only part of the field or is indirect. Mixed: the evidence is split, or roughly half express indifference to or acceptance of building such systems. Mostly false: a substantial plurality say they are willing to build AI they consider likely misaligned or harmful. False: a majority of AI researchers say they want or intend to build AI that is harmful to humans.

Evidence

  1. 2022 Expert Survey on Progress in AI [AI Impacts Wiki] AI Impacts · 4 August 2022
    “69% of respondents think society should prioritize AI safety research more or much more, up from 49% in 2016.”
  2. [PDF] Ethics and Governance of Artificial Intelligence: A Survey of Machine Learning Researchers (Extended Abstract) IJCAI-22 (authors from Syracuse University, Centre for the Governance of AI, Council on Foreign Relations, University of Pennsylvania, DeepMind) · 2022
    “When asked about prioritizing AI safety an overwhelming majority of our respondents (68%) say that the field should be prioritized more than at present.”
  3. Thousands of AI Authors on the Future of AI arXiv (Grace et al.; AI Impacts)
    “A large majority of respondents thought that AI safety research should be prioritized more than it currently is.”
  4. [PDF] Ethics and Governance of Artificial Intelligence: A Survey of Machine Learning Researchers (Extended Abstract) IJCAI-22 (authors from Syracuse University, Centre for the Governance of AI, Council on Foreign Relations, University of Pennsylvania, DeepMind) · 2022
    “31% of researchers indicate that they would resign or threaten to resign from their jobs, and 25% indicate that they would speak out publicly to the media or online, if their organization decided to work on lethal autonomous weapons.”

Reasoning

Independent surveys of AI researchers show that clear majorities want AI safety research prioritised more. The 2022 AI Impacts survey puts this at 69% [E1], an ML researcher survey at 68% [E2], and the later survey of thousands of AI authors calls it "a large majority" [E3]. A meaningful share of ML researchers also say they would resign or speak out if their organisation worked on lethal autonomous weapons [E4]. This points to a majority preferring safe, beneficial outcomes. But none of the surveys asks directly whether researchers want to build misaligned or harmful AI, so the evidence is indirect, which fits "mostly true" under the criteria.

Cite this claim

Eliezer Yudkowsky, “Most AI researchers will not want to make unFriendly AIs.” (said 2008). Verdict: mostly true. Reliability API, claim c_f7fa4f6e3ac3, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_f7fa4f6e3ac3

Claims without a verdict

Is a verdict here wrong, or do you speak for Eliezer Yudkowsky? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.