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
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
Not researched Factual 2008 Yudkowsky asks readers to consider how many media stories on artificial intelligence cite the Terminator movies as if they were documentaries, and how many media stories on brain-computer interfaces mention Star Trek's Borg. Factual · 2008 Their own work or organisation
Not checkable Not checkable 2008 It’s mistaken belief that an AI will be friendly which implies an obvious path to global catastrophe. Not checkable · 2008
Not checkable Not checkable 2008 In retrospect this is what one would expect. Not checkable · 2008
Not checkable Not checkable 2008 There should be a curriculum for people like ourselves; a list of skills we need in addition to our domain-specific knowledge. Not checkable · 2008
Not checkable Not checkable 2008 Intelligence, to be useful, must be used for something other than defeating itself. Not checkable · 2008
Not checkable Not checkable 2008 Technical failure is when you try to build an AI and it doesn’t work the way you think it does Not checkable · 2008
Not checkable Not checkable 2008 The lesson of history is that swan happens. Not checkable · 2008
Not researched Factual 2008 Homo sapiens represents the first general intelligence. 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
Not checkable Not checkable 2008 To survive any appreciable time, we need to drive down each risk to nearly zero. Not checkable · 2008
Not researched Factual 2008 The human brain cannot release enough neurotransmitters to feel emotion a thousand times as strong as the grief of one funeral. Factual · 2008 Philosophy, ethics and society
Not checkable Not checkable 2008 To hit such a tiny target in configuration space requires a powerful optimization process. Not checkable · 2008
Not checkable Not checkable 2008 Friendly AI is not a module you can instantly invent at the exact moment when it is first needed, and then bolt on to an existing, polished… Not checkable · 2008
Not checkable Not checkable 2008 If the augment is substantially smarter than human, this too is a global catastrophic risk. Not checkable · 2008
Not checkable Not checkable 2008 You cannot "rationalize" what is not rational to begin with - as if lying were called Not checkable · 2008
Not checkable Not checkable 2008 According to probability theory, adding additional detail onto a story must render the story less probable. Not checkable · 2008
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
Not researched Factual 2008 People deny being anchored or contaminated even when experiments show they are. Factual · 2008 Science and research methods
Not checkable Not checkable 2008 The answer is that I don’t know because I don’t have eons to ponder. Not checkable · 2008
Not checkable Not checkable 2008 Even if you abandoned all thoughts of fun, gave up living to preserve your life, you couldn’t navigate a million-year obstacle course. Not checkable · 2008
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
Not checkable Not checkable 2008 Not every change is an improvement, but every improvement is necessarily a change. Not checkable · 2008
Not checkable Not checkable 2008 One path leading to global catastrophe—to someone pressing the button with a mistaken idea of what the button does—is that Artificial… Not checkable · 2008
Not checkable Not checkable 2007 Writing a great deal of exploratory code means that IDEs with refactoring support and version control are important, and that modular code… Not checkable · 2007
Not checkable Not checkable 2007 Programmers may try to second-guess isolation by teaching many similar knowledges at around the same time, but that is no substitute for a… Not checkable · 2007
Not checkable Not checkable 2007 Clustering on a single feature is definitely not a complete conceptual system. Not checkable · 2007
Not researched Factual 2007 DGI hypothesizes that mental visualization occurs through a multistage flow from high-level feature controllers to low-level feature controllers within a sensory modality. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 Brainware is limited to what the programmer was thinking about at the time. Not checkable · 2007
Not checkable Not checkable 2007 Concepts are associated with other concepts. Not checkable · 2007
Not checkable Not checkable 2007 in which victory is never declared, and even after a problem appears to be solved, we go on trying to solve it. Not checkable · 2007
Not checkable Not checkable 2007 Similarly, the abstract knowledge that biological neurons implement human intelligence does not explain human intelligence. Not checkable · 2007
Not checkable Not checkable 2007 Evolution is a good teacher but a poor role model Not checkable · 2007
Not checkable Not checkable 2007 I would guess not Not checkable · 2007
Not researched Factual 2007 The easiest, most evolvable neural architecture is generally one in which the feedback connection reciprocates the feedforward connection. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 Thoughts are the disposable one-time structures implementing a nonrecurrent mind in a non-recurrent world. Not checkable · 2007
Not researched Citation 2007 Mervis, Catlin and Rosch (1975) found that distinguishing colors across innate color boundaries is easy while distinguishing within boundaries is hard. Citation · 2007 Science and research methods
Not checkable Not checkable 2007 Programmers are trained to solve problems, and trying to create general AI means solving the programming problem of creating a mind that… Not checkable · 2007
Not checkable Not checkable 2007 Imagery supports abstract percepts. Not checkable · 2007
Not researched Factual 2007 Human thoughts, the source of human-written code, are resilient while the code itself is fragile. Factual · 2007 AI capabilities
Not researched Factual 2007 In most human-written code, a small perturbation of any kind usually causes a crash. Factual · 2007 AI capabilities
Not researched Factual 2007 Yudkowsky wrote in 2007 that he strongly doubted that human cognitive systems create concepts by training generic neural nets on satisfaction and imposition. Factual · 2007 Philosophy, ethics and society
Not researched Factual 2007 Humans are not preprogrammed by evolution with knowledge about butterflies. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 shifting to microworlds does require that experience in the microworlds reproduce functionally relevant aspects of experience in real life Not checkable · 2007
Not checkable Not checkable 2007 The challenge is using new advantages to compensate for the loss of old advantages, and replacing substrate-level support with design-level… Not checkable · 2007
Not checkable Not checkable 2007 The first principle is that of multiple levels of organization. Not checkable · 2007
Not checkable Not checkable 2007 The best way to ensure that the AI can modify and use internal content is to have the AI create the content. Not checkable · 2007
Not checkable Not checkable 2007 Even in the absence of an immediate environmental regularity, a concept can contribute to effective intelligence by enabling the perception… Not checkable · 2007
Not checkable Not checkable 2007 Designing a modality that worked equally well for any possible computer program would probably be an indication that the modality was… Not checkable · 2007
Not checkable Not checkable 2007 but I do not believe that the rooster crowing causes the sun to rise. Not checkable · 2007
Not checkable Not checkable 2007 I believe that rather than simplicity, we should pursue Not checkable · 2007
Unresolvable Factual 2007 Humans were created by evolution, an unintelligent process, whereas AI will be created by intelligent human processes. Factual · 2007 AI capabilities
Not researched Factual 2007 The human visual modality precategorizes colors into a color space with eleven culturally universal focal volumes that have comparatively sharp boundaries. Factual · 2007 Philosophy, ethics and society
Not checkable Not checkable 2007 The option containing five billiards, regardless of their size or relative positions or movement patterns, is the key to the gate. Not checkable · 2007
Not checkable Not checkable 2007 An attempt to blindly imitate the human visual modality, without understanding the role of modalities in intelligence, would be unlikely to… Not checkable · 2007
Not checkable Not checkable 2007 If the “knob” was created on the fly, it implies a much higher degree of systemic creativity than selecting from among pre-existing options. Not checkable · 2007
Not researched Factual 2007 The human brain makes no distinction between CPU and RAM. Factual · 2007 AI compute, hardware and data
Not checkable Not checkable 2007 Simplicity is the grail of physics, not AI. Not checkable · 2007
Not checkable Not checkable 2007 Note that primitive does not mean architecturally simple. Not checkable · 2007
Not checkable Not checkable 2007 Cleanup should be undertaken, not with pride in the greater simplicity of human design relative to evolutionary design, but with a healthy… Not checkable · 2007
Not checkable Not checkable 2007 Intelligence is a supersystem composed of many mutually interdependent subsystems—subsystems specialized not only for particular… Not checkable · 2007
Not checkable Not checkable 2007 a human programmer wields general intelligence like a scalpel. Not checkable · 2007
Not researched Factual 2007 At a limiting speed of 150 m/s for myelinated axons, each human neuron is potentially within roughly one clock tick of any other neuron in the brain. Factual · 2007 Science and research methods
Not researched Citation 2007 Lakoff (1987) shows that the classical necessary-and-sufficient-requisites view of concepts is incompatible with a wide range of established cognitive effects. Citation · 2007 Science and research methods
Not checkable Not checkable 2007 merely “relatively large”) number of variations on the underlying challenge—enough variations to prevent the AI from solving the problem… Not checkable · 2007
Not researched Factual 2007 Reacting negatively to a fleeting mental image of sharp glass requires no extended line of reasoning and can happen in a single perceptual step. Factual · 2007 Philosophy, ethics and society
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
Not checkable Not checkable 2007 I recommend eliminating environmental reconstruction as a complexity source in early AI development. Not checkable · 2007
Not researched Factual 2007 The feature controllers used to create mental imagery are evolvable and preadaptive in the absence of mental imagery. Factual · 2007 Science and research methods
Not researched Factual 2007 As of 2007 there is no explanatory consensus for the existence of the massive corticothalamic feedback projections. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 Abandoning the human modalities appears to introduce an additional fragile dependency on the correctness of the AI theory Not checkable · 2007
Not checkable Not checkable 2007 We might call such concepts surface concepts. Not checkable · 2007
Not checkable Not checkable 2007 In AI the rule should be that the problem is always harder and deeper than it looks, even after you take this rule into account. Not checkable · 2007
Not checkable Not checkable 2007 But on the whole, the modality level is about invariants rather than regularities and identities rather than categories. Not checkable · 2007
Not checkable Not checkable 2007 category structure is ubiquitous in the low-entropy processes of our low-entropy universe. Not checkable · 2007
Not checkable Not checkable 2007 I find it likely that the entire operation will cancel out, contributing nothing. Not checkable · 2007
Not checkable Not checkable 2007 “Fuzzy hashing” would then be a strong candidate for an alternative implementation of memory association. Not checkable · 2007
Not checkable Not checkable 2007 This idiom does not work for general AI! Not checkable · 2007
Not checkable Not checkable 2007 Concepts can also embody regularities which are not directly represented in any sensory modality, and which are not any covariance or… Not checkable · 2007
Not researched Factual 2007 Evolution has produced a general intelligence (the human brain) running on about a hundred trillion synapses with limiting speeds of roughly 200 spikes per second. Factual · 2007 Science and research methods
Not checkable Not checkable 2007 This is mathematically straightforward, but cognitively general. Not checkable · 2007
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
Not checkable Not checkable 2007 Semantic nets leave out multiple levels of organization. Not checkable · 2007
Not checkable Not checkable 2007 Abstract reasoning is a means by which inductively acquired generalizations can be used in deductive reasoning. Not checkable · 2007
Not checkable Not checkable 2007 In a deliberative AI, it is patterns of thoughts that plan and design, transforming abstract high-level goal patterns into specific… Not checkable · 2007
Not researched Factual 2007 Visualizing a "red lemon" does not cause the human mind to picture a bright red patch with the outline of a lemon. Factual · 2007 Philosophy, ethics and society
Not checkable Not checkable 2007 Words are the phonemic tags (speech), visual tags (writing), gestural tags (sign language), or haptic tags (Braille) used to invoke… Not checkable · 2007
Not checkable Not checkable 2007 The gap between the higher level and the lower level is not absolute and uncrossable, as it is in semantic nets. Not checkable · 2007
Not checkable Not checkable 2007 A concept-formation system need not be generally intelligent in itself Not checkable · 2007
Not checkable Not checkable 2007 but this severs the task of concept formation from its ecological validity (metaphorically speaking). Not checkable · 2007
Not researched Factual 2007 Many proverbial AI pathologies result at least partially from omitting lower levels of organization from the design. Factual · 2007 AI capabilities
Not checkable Not checkable 2007 Neural networks and evolutionary computations are not generally intelligent but they are generically intelligent; Not checkable · 2007
Not checkable Not checkable 2007 The visual cortex does not know about butterflies. Not checkable · 2007
Not researched Citation 2007 Judd et al. (1964) measured 622 spectral power distributions of natural lighting and found that variations in natural lighting reduce to three degrees of freedom. Citation · 2007 Science and research methods
Not checkable Not checkable 2007 Concepts also have complexity that relates to the concept level Not checkable · 2007
Not checkable Not checkable 2007 A piece of code need only evolve to test for the same characteristic. Not checkable · 2007
Not researched Factual 2007 It is easier to train neural networks, including nonbiological connectionist networks, than to train a piece of human-written code. Factual · 2007 AI capabilities
Not researched Citation 2007 Evidence from neuroanatomy, neuroimaging, neurological pathology and cognitive psychology (Kosslyn 1994) supports that mental imagery is directly represented in sensory modalities. Citation · 2007 Science and research methods
Not checkable Not checkable 2007 I believe that many (though not all) cases of the “commonsense problem” or “frame problem” arise from trying to store all possible… Not checkable · 2007
Back to the list

Not checkable

“but this severs the task of concept formation from its ecological validity (metaphorically speaking).”

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

but this severs the task of concept formation from its ecological validity (metaphorically speaking).

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

Cite this claim

Eliezer Yudkowsky, “but this severs the task of concept formation from its ecological validity (metaphorically speaking).” (said 2007). Verdict: not checkable. Reliability API, claim c_39e3e38b627e, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_39e3e38b627e

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.