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

2,016 claims, sorted by newest said. Choose a claim to read it in full.
VerdictKindSaidClaimTopic
Not checkable Not checkable 2007 Evolution is a useful inspiration but a dangerous template. Not checkable · 2007
Not checkable Not checkable 2007 Concepts reside in long-term storage; thoughts affect specific imagery. Not checkable · 2007
Not checkable Not checkable 2007 The archetypal examples of “thoughts” (invented, specific, instantiated, non-recurrent) are the sentences mentally “spoken” and mentally… Not checkable · 2007
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
Not checkable Not checkable 2007 Thoughts are not evolved/programmed brainware, or a long-term reservoir of learned complexity; thoughts are constructed on a… Not checkable · 2007
Mixed Factual 2007 Smooth fitness landscapes are a necessary and fundamental outcome of evolution. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 Forming a complex concept requires an incremental path to that complex concept—a series of building-block concepts and precursor concepts… Not checkable · 2007
Not checkable Not checkable 2007 A concept is abstracted from experiences that exist as sensory patterns in one or more modalities. Not checkable · 2007
Not checkable Not checkable 2007 I emphasize that the exercise is not intended as evidence for a theory, but rather as a means of introducing and grounding concepts that… Not checkable · 2007
Not checkable Not checkable 2007 Once abstracted, a concept can be compared to a new sensory experience to determine whether the new experience satisfies the concept Not checkable · 2007
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
Not checkable Not checkable 2007 Unique correspondence, as a mental skill, tends to be reinforced by any goal-oriented challenge in which a useful object cannot be in two… Not checkable · 2007
Not researched Factual 2007 The human brain's most fundamental limit is its speed, with under 200 sequential operations available for anything taking less than a second. Factual · 2007 Science and research methods
Not researched Factual 2007 Yudkowsky wrote that he suspects a concept kernel consists of a pattern in a representation related to (but not identical with) the representation of sensory imagery, and that this pattern is produced by transforming the experiences from which the concept is abstracted. Factual · 2007 Philosophy, ethics and society
Not checkable Not checkable 2007 this smooth reversibility helps render it possible to learn a single concept which can act as a category detector or a category imposer. Not checkable · 2007
Not checkable Not checkable 2007 The problem with the classical view is that it allows only a limited repertoire of concepts. Not checkable · 2007
Not researched Factual 2007 In humans, the concept of red is probably at the surface of the visual modality, a direct output of its feature-detectors. Factual · 2007 Philosophy, ethics and society
Not checkable Not checkable 2007 Kernels relate concepts to sensory imagery and hence the modality level. Not checkable · 2007
Not checkable Not checkable 2007 Even if an AI needs programmer-created concepts to bootstrap further concept formation, bootstrap concepts should be created using… Not checkable · 2007
Not checkable Not checkable 2007 Today we have at least one reason to believe that nonsensory intelligence is a bad approach; we tried it and it didn’t work. Not checkable · 2007
Not checkable Not checkable 2007 An arithmetic expert system that leaves out one level of organization may be recoverable. Not checkable · 2007
Not checkable Not checkable 2007 Concepts possess transformation relations, which are again illustrated in Copycat. Not checkable · 2007
Not checkable Not checkable 2007 is the pseudo-sensory pattern produced by abstracting from sensory experience. Not checkable · 2007
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
Not checkable Not checkable 2007 Concepts with procedural definitions are regular concepts and may possess kernels, exemplars, associated memories, and so on. Not checkable · 2007
Not researched Factual 2007 Yudkowsky stated that he did not believe concept-concept relations are computed by directly comparing kernel representations, and that he thought they are learned by generalizing across the concepts' usage. Factual · 2007 Philosophy, ethics and society
Not checkable Not checkable 2007 In fact, a visual cortex “knows” nothing; a sensory modality contains behaviors which correspond to environmental invariants, not knowledge… Not checkable · 2007
Not checkable Not checkable 2007 A human-programmed numeron detector, mutated on the code level, would probably simply break. Not checkable · 2007
Not checkable Not checkable 2007 Once an AI supersystem possesses any degree of intelligence at all, no matter how primitive, that intelligence becomes a tool which can be… Not checkable · 2007
Not checkable Not checkable 2007 Effective intelligence is produced not by having concepts but by using concepts; one learns to use concepts by using them. Not checkable · 2007
Not checkable Not checkable 2007 Most of the computing power is required to find the hint, not to verify the match, and a false hint does no damage Not checkable · 2007
Not checkable Not checkable 2007 For a Newtonian billiards modality, an individual “gate” (subtask) might consist of four “option systems,” each option system grouped into… Not checkable · 2007
Mostly true Factual 2007 Under GOFAI and formal logic, the classical view of concepts treats abstraction as information loss. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 This essential fact about the structure of our low-entropy universe is what makes intelligence possible, computationally tractable… Not checkable · 2007
Not checkable Not checkable 2007 The successive layers of organization in a sensory modality are a beautiful illustration of evolution’s design signature Not checkable · 2007
Not checkable Not checkable 2007 this is like talking about the United States by referring to New York. Not checkable · 2007
Not checkable Not checkable 2007 Holonic conflict resolution is not proposed as a complete solution to perceptual problems Not checkable · 2007
Not checkable Not checkable 2007 Thus, I would attribute the symbol grounding problem to “collapsing the levels of organization,” rather than “hardwiring what should be… Not checkable · 2007
Not checkable Not checkable 2007 Intelligence in the fully human sense requires the ability to manipulate the world by reasoning backward from a mental image of the desired… Not checkable · 2007
Not checkable Not checkable 2007 certain specific complexity is required, and complexity for the sake of complexity is worse than useless. Not checkable · 2007
Not checkable Not checkable 2007 Cognition goes beyond passive denotation; it can predict future sensory data from past experience. Not checkable · 2007
Not checkable Not checkable 2007 I cannot rule out the possibility that a more complex, more deeply creative process was involved in the transition from triangle to pyramid Not checkable · 2007
Not checkable Not checkable 2007 Even so, the complexity involved in the redness concept lies almost entirely within the sensory modality, rather than the concept kernel. Not checkable · 2007
Not checkable Not checkable 2007 Concepts are faceted; they have internal structure and associational structure which comes into play when imposition or description… Not checkable · 2007
Not checkable Not checkable 2007 Placing virtual pegs in virtual holes is admittedly not an inherently impressive result. Not checkable · 2007
Not checkable Not checkable 2007 Concepts act as verbs, adjectives, and adverbs as well as nouns. Not checkable · 2007
Not checkable Not checkable 2007 The fundamental reason for preferring a nonmodality solution in this example is to demonstrate that an AI must be capable of solving… Not checkable · 2007
Not checkable Not checkable 2007 I would suggest using constant-resolution sensory modalities during the early stages of an AI Not checkable · 2007
Not researched Factual 2007 Yudkowsky said he doubted that feature controllers reach all the way down to the first layers of the retina, noting that he had not heard of any feedback connections reaching that far. Factual · 2007 Science and research methods
Not researched Factual 2007 Human-written programs typically involve long chains of dependencies that intersect at single points of failure. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 Programming inherently violates the levels of organization; it leads directly into the pitfalls of classical AI. Not checkable · 2007
Not checkable Not checkable 2007 Even simultaneously satisfying the abstraction and satisfaction problems Not checkable · 2007
Not checkable Not checkable 2007 Concepts are patterns that mesh with sensory imagery; concepts are complex, flexible, reusable patterns that have been reified and placed… Not checkable · 2007
Not checkable Not checkable 2007 It is acceptable for the discovery of “fiveness” to be straightforward, but the discovery method must be general. Not checkable · 2007
Not checkable Not checkable 2007 This illustrates the second-worst error in AI, the failure to distinguish between things that can be hardwired and things that must be… Not checkable · 2007
Not checkable Not checkable 2007 I would advocate placing the AI in complex virtual environments, where the virtual environments are internal to the computer but external… Not checkable · 2007
Not checkable Not checkable 2007 I would argue that it is generally a feature. Not checkable · 2007
Not checkable Not checkable 2007 I would argue that the existence of an additional dependency is illusory. Not checkable · 2007
Not checkable Not checkable 2007 The programmers would then attempt to develop sensory modalities corresponding to the virtual environments. Not checkable · 2007
Not checkable Not checkable 2007 Omitting all the experiential and sensory grounding of human symbols leaves no raw material to work with. Not checkable · 2007
Not checkable Not checkable 2007 There is an immediately obvious way to train a neural net to detect satisfaction of Not checkable · 2007
Not checkable Not checkable 2007 constructing complex mental imagery requires that a concept target a piece of existing mental imagery, which the concept then transforms. Not checkable · 2007
Not researched Factual 2007 Each ascending layer of the code-modality-concept-thought-deliberation model of the human mind is more elaborate, complex, flexible and computationally expensive. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 Rather than fiveness, the AI would have formed the concept “things-it-takes-about-20-milliseconds-to-understand.” Not checkable · 2007
Not researched Factual 2007 Intelligence in the human cognitive supersystem results from many cognitive processes taking place on multiple levels of organization. Factual · 2007 Philosophy, ethics and society
Not checkable Not checkable 2007 Concepts can describe both environmental sensory experience and internally generated mental imagery. Not checkable · 2007
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
Not researched Factual 2007 Evolution uses existing genetic complexity to build new genetic complexity but exhibits no foresight. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 The short, snappy auditory tag “five” can stand in for the complexity bound up in the fiveness concept. Not checkable · 2007
Not checkable Not checkable 2007 For the purposes of this discussion, we are not speaking of the ability to understand numbers, arithmetic, or mathematics, only an AI’s… Not checkable · 2007
Not checkable Not checkable 2007 Humans are not intelligent by virtue of being “supersystems,” but by virtue of being a particular supersystem which implements human… Not checkable · 2007
Not checkable Not checkable 2007 a human designer can use foresight to plan multiple new system components as part of a coordinated upgrade. Not checkable · 2007
Not researched Factual 2007 Copycat has a direct procedural implementation of number sense and does not use mapping constructs to build numerical concepts. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 Each option would have a different configuration; the AI would choose one of the four options based on its current hypothesis about the… Not checkable · 2007
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
Not checkable Not checkable 2007 The lesson I would learn out from this morality tale of color constancy is that sensory modalities are about invariants and not just… Not checkable · 2007
Unresolvable Factual 2007 The idea of 'pure' abstract reasoning has historically given rise to pathologies in AI research. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 A concept with internal structure or procedural structure, created by the AI’s own thought processes in response to experience, is mutable… Not checkable · 2007
Not checkable Not checkable 2007 Possession of the category “five” does not even imply the possession of the categories “four” or “six,” much less the formulation of the… Not checkable · 2007
Not checkable Not checkable 2007 The target matter of Artificial Intelligence is not the surface variation that makes one human slightly smarter than another human, but… Not checkable · 2007
Not checkable Not checkable 2007 a design based only on consensus contemporary theory might omit feature controllers! Not checkable · 2007
Not checkable Not checkable 2007 A real AI would be the ultimate general solution because it would encapsulate the cognitive processes that human programmers use to write… Not checkable · 2007
Not checkable Not checkable 2007 are emergent from the underlying representation of concepts Not checkable · 2007
Not checkable Not checkable 2007 The regularities in an “expert system for arithmetic” would be real, noticeable by simple and computationally feasible means, and could be… Not checkable · 2007
Not researched Factual 2007 Feature controller layers in the brain are not a separate system independent from the feature detector layers. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 The concrete theory presented in Section 2 goes under the name of “deliberative general intelligence” (DGI). Not checkable · 2007
Not checkable Not checkable 2007 I believe that the search for a single essence of intelligence lies at the center of AI’s previous failures. Not checkable · 2007
Not checkable Not checkable 2007 I suspect it will usually be best to separate out a less expensive cueing mechanism as the background process Not checkable · 2007
Not checkable Not checkable 2007 to solve all three problems simultaneously constitutes the fundamental challenge of building a system that learns complexity in… Not checkable · 2007
Not checkable Not checkable 2007 If all the LISP tokens in a semantic net were given random new names, there would be no way to deduce whether G0025 formerly meant… Not checkable · 2007
Not checkable Not checkable 2007 A semantic net not only lacks richness, it lacks the capacity to represent richness. Not checkable · 2007
Not researched Factual 2007 Yudkowsky wrote that Lenat tried self-optimizing heuristics (Eurisko), which failed to yield intelligence, and then moved on to Cyc. He framed this as a way of understanding the field's behaviour if one postulates that Eurisko was treated as a failed hypothesis. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 I believe that the most important skill an AI developer can have is knowing what not to program. Not checkable · 2007
Not checkable Not checkable 2007 this cognitive superprocess will hereafter be referred to as deliberation. Not checkable · 2007
Not researched Factual 2007 Human cognition may be massively parallel at lower levels, but the overall flow of cognition is serial. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 but that is no excuse for ignoring concept-concept relations in a new theory. Not checkable · 2007
Not checkable Not checkable 2007 Isolation means that if a concept, or a piece of knowledge, is handed to the AI on a silver platter, the AI may be isolated from the things… Not checkable · 2007
Not checkable Not checkable 2007 The correspondence happens on the level where 13 is the structure 00001101, not on the level where it is the number thirteen. Not checkable · 2007
Not checkable Not checkable 2007 the possession or creation of this knob is the most interesting part of the process, not the selection of one knob from many. Not checkable · 2007
Not checkable Not checkable 2007 We already know what the ultimate bottom layer of an Artificial Intelligence looks like; it looks like ones and zeroes. Not checkable · 2007
Back to the list

Not checkable

“Evolution is a useful inspiration but a dangerous template.”

Eliezer Yudkowsky 2007 Levels of Organization in General Intelligence (MIRI)

Evolution is a useful inspiration but a dangerous template.

Not checkable An opinion, value judgement or statement with nothing to check.

Cite this claim

Eliezer Yudkowsky, “Evolution is a useful inspiration but a dangerous template.” (said 2007). Verdict: not checkable. Reliability API, claim c_9a0e057e349e, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_9a0e057e349e

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.