What this covers
Michelle Dawson speaks with Tyler Cowen about how the scientific and clinical understanding of autism has gone systematically wrong, and what correcting that misunderstanding would require. Dawson, a researcher who has made major contributions to autism science while being autistic herself, argues that the field confuses atypicality with deficiency—treating information-processing differences as cognitive failures rather than cognitive differences. The conversation moves across psychology, medicine, law, and education, with Dawson drawing on her own research and that of colleagues to challenge both the deficit model and the field's institutional standards.
The discussion opens onto several specific domains where autistic performance contradicts the standard deficit narrative. Raven's Progressive Matrices, a test of abstract reasoning, reveals that autistics often perform substantially better than on IQ batteries that penalize their processing style—a finding the field has historically dismissed by questioning the test rather than revising its theory. Dawson traces parallel reversals in music perception (autistics excel at chord disembedding, famously demonstrated by Derek Paravicini), reading (hyperlexia as a genuine early strength), and resistance to cognitive biases like framing effects. She argues these are not scattered savant talents but expressions of a more general cognitive versatility—autistic brains process information more optionally, without the mandatory hierarchies that typical brains impose to make information manageable.
Dawson also takes up systemic issues: the harm caused by interventions that ration information away from autistic people, the logical fallacy of conflating atypicality with severity, the troubling finding that some autistics diagnosed late or not at all had better outcomes, and the uniquely low standards of science and ethics the autism field accepts without professional resistance. She distinguishes between the claims made by influential advocates and what the literature actually supports, and discusses why older, undiagnosed autistics have been effectively erased from research.
Dawson argues that autism is best understood not as a deficit but as domain-general atypical information processing marked by cognitive versatility and less mandatory hierarchies, and that the field's low scientific and ethical standards—rather than autism itself—systematically harm autistic people.
- Autistic brains process information more optionally/autonomously, attenuating biases and enabling abilities long dismissed as dead ends
- Common interventions ration information away from autistics, which damages their development of abilities
- An autism diagnosis effectively segregates people into lower standards of science and ethics than would be acceptable for anyone else
Autism diagnosis enforces unacceptably low standards of science and ethics not applied to other conditions.
- A DSM autism diagnosis is itself a form of segregation because it relegates autistic people to lower standards of science and ethics than would be acceptable for anyone else; getting good standards becomes a matter of luck (encountering exceptional individuals) rather than a default with recourse if denied.
“DSM autism is segregation. Why? Because the diagnosis segregates autistic people into lower standards of science and ethics than would be acceptable for anybody else.”
- Autism is uniquely positioned among diagnoses in having unacceptably low standards of science and ethics that are universally advocated for, with essentially no advocacy in the opposite direction, so that the influence of advocacy is completely determinative — a situation Dawson says she has not found in other diagnoses.
“you have these unacceptably low standards that are universally advocated for, and that there was no advocacy in the other direction”
- What helped gay individuals was gaining human rights and the protection of basic standards in science and ethics, not identity politics; therefore what autistics need is those rights and standards rather than identity politics.
“what helped gay individuals is having human rights and the benefit and protection of basic standards in science and ethics. I think that's what autistics are missing, not identity politics.”
- One of the most common features of popular autism interventions is to ration information and keep it away from autistic people, which is harmful because autistic abilities develop through access to the information they process well; treating these abilities as dead ends is mistaken.
“one of the most common aspects of popular interventions is to really ration information, keep it away from autistic people”
- Dawson learned autism research out of dire personal and legal necessity by directly reading large volumes of primary papers — and associated materials like trial registrations and grant information — despite being repeatedly told she was too stupid to understand the literature, finding that the information she needed was in those papers.
“Because I had to address these legal issues and questions, I did actually look at the autism literature, and suddenly I had information I could really work with.”
- Firms like Specialisterne that test, place, and coach autistic workers effectively give them an 'endorsement' or stamp of approval that signals to employers how to treat them, making their employment far less precarious than going it alone where they risk being denied basic rights and standards regardless of how good their work is — though it is a form of segregation that ideally should not be necessary.
“you would be wise to seek out enterprises like this because they are... First off, obviously, they shouldn't be necessary.”
Autistic brains process information atypically across all domains without mandatory hierarchies or cognitive editing.
- The best understanding of autism is atypical brain functioning that results in atypical processing of ALL information across domains (social and nonsocial), modalities (visual, auditory), and sources (memory or external world) — a very domain-general atypicality.
“The best understanding is seeing autism as atypical brain functioning, resulting in atypical processing of all information.”
- Autistic people can process more information across levels and scales without losing it, without the cognitive editing that typical brains use to make information manageable, and can still learn from and use that unedited information.
“an autistic person can process more information across levels and scales without losing it, or without a lot of the information being edited in sort of cognitive editing”
- It is well established that autistics do not organize information in memory typically and do not use narratives to edit or organize it, which explains their different performance on episodic memory tasks and tests requiring organization into common narratives.
“autistics do not organize information in memory typically”
- Autistic brains' atypicality involves cognitive versatility and less mandatory hierarchies: they consider more possibilities and nonstrategically combine information across levels and scales without losing large parts of it, applying this to all information.
“an autistic person can process more information across levels and scales without losing it, or without a lot of the information being edited in sort of cognitive editing”
- Because typical people share a similar information-processing hierarchy (giving them much in common despite different genes and experiences) while autistics lack this shared hierarchy, autistics are not only different from typical people but quite different from each other — increasingly recognized as autistic idiosyncrasy.
“Autistics do not have that, which makes them not just different from typical people, but quite different from each other.”
Autistics demonstrate genuine cognitive strengths in processing, reasoning, and perceptual discrimination that tests routinely miss.
- In a categorization study (Soulières), autistics formed perfectly typical category boundaries when classifying ellipses but, when asked to discriminate, did not show the typical discrimination peak — meaning discrimination was autonomous from categorization; for typical people categorization is mandatory even when it biases and is not useful, whereas for autistics top-down processing/priors are optional and versatile rather than automatic.
“In the typical population, even when it wasn't useful, they formed categories — they had to.”
- Autistics are good at chord disembedding (disaggregating a chord into its component notes); originally established via autistic savants, it was confirmed in less obviously talented autistics, with savant Derek Paravicini able to disembed even a 10-note chord — the researchers ran out of fingers before reaching his limit.
“Autistics are good at chord disembedding, and that's what that ability is called.”
- Hyperlexia is a classic early autistic strength: in book-rich environments with free access, autistic children as young as two spontaneously figure out the structure of even difficult orthographies like English, working out irregularities across letters, words, phrases and whole texts — something typical children do not do spontaneously.
“The almost classic example is hyperlexia, which you see starting with very early development.”
- When a 2004 paper tentatively found autistics doing better on Raven, the field interpreted it as Raven being a bad test that overestimated autistic intelligence; Dawson instead pursued the alternative that the finding was real, noting the autism field's pattern of concluding that if autistics do well on a test, something must be wrong with the test.
“The interpretation in the paper was that Raven wasn't a good test for autistics. It must overestimate autistic intelligence.”
- Autistics perform substantially better on Raven's Progressive Matrices than on Wechsler-type batteries, sometimes spectacularly so, even though Raven demands exactly the complex processing, executive function and working memory abilities autistics are assumed to lack; this challenges deficit-based interpretations of autistic intelligence.
“Autistics do better on Raven, quite a bit better. In some individuals, the discrepancy is spectacular”
Society's bias that atypicality signals lower ability obscures autistic strengths and harms both people and research.
- Societies (more at the collective/median level than individually) are very bad at detecting abilities and worth in highly atypical individuals, defaulting to the assumption that the more atypical a person is, the worse and less intelligent they must be — a bias that has both harmed autistic people and impeded research.
“it seems to be something that people are very bad at, to detect abilities and worth in highly atypical individuals”
- There is no workable definition of autism severity; the term is used loosely to mean a bad outcome, bad behavior, or a low IQ score, encoding the unhelpful bias that the more atypical a person is, the more must be wrong with them.
“There is no workable definition of autism severity.”
- There are no confirmed biases that are stronger in autistics; because autistics process information atypically and non-mandatorily consider other possibilities rather than defaulting to the typical hierarchy, biases tend to be attenuated rather than strengthened — illustrated by autistics drawing impossible figures well yet still knowing the figures are impossible.
“the autistics are less likely to be fooled by the framing. So, they detect when two situations are the same”
Early autism diagnosis may be harmful; autistics differ markedly from each other and have underestimated reproductive fitness.
- Contrary to prevailing advice favoring early diagnosis, purpose-built and population-based studies (including one in Korea) found that probable autistics who were not diagnosed, or who were diagnosed at older rather than younger ages, were often better off — raising the hard, under-investigated question of whether an autism diagnosis is more harmful than beneficial.
“the autistics who had not been diagnosed — did not have an existing diagnosis — were much better off.”
- There are many autistic parents of autistic offspring, a poorly studied phenomenon, and the reproductive fitness of autistics has been wildly underestimated.
“The reproductive fitness, as they call it, of autistics has been wildly underestimated, I think.”
- Autism has not become more prevalent over time and there is no reason to believe it has; the belief that prevalence increased is destructive because it effectively erases older autistics, denying researchers the chance to learn from their development, outcomes, and achievements.
“autism has not become more prevalent, and there is no reason to believe that it has”
- Autism's loud, influential advocates — autistic and nonautistic, professional organizations, service sellers — are not representative of parents or autistics generally; the scientific literature differs from what loud advocates claim, and one cannot extrapolate from advocacy to the actual interests or views of any group like parents.
“you have very, very loud people — loud advocates, loud influential advocates... They are not necessarily representative.”
Autistics resist false social inferences not from social inability but from rational caution about inferring mental states from limited data.
- The most replicated autistic deficit is attributing deep mental states (e.g. jealousy, flirting) to animated geometric shapes; Dawson interprets this not as failure to see the social resemblance but as autistics refusing to discard the accurate information that these are 'just triangles,' producing reasonable caution about inferring profound mental states.
“This is the task of attributing deep mental states to geometric shapes. Usually they are triangles.”
- The autism literature on judging facial expressions caricatures both autistics and the typical population, falsely assuming emotions are simply readable from a face; the broader literature is more complex (e.g. people produce genuine Ekman-type smiles when frustrated), and training autistics to read simplified, stereotyped expressions teaches them to ignore the real, complex, important information.
“you haven't only just turned autistic people into stereotypes and cartoons, you've done that to the typical population”
- Autistics are less likely to be fooled by framing effects: they more often notice that two differently-posed offers yield identical outcomes, a finding replicated once, whereas the typical group is biased into being fooled by the framing.
“the autistics are less likely to be fooled by the framing. So, they detect when two situations are the same”
Savant abilities cluster in domains with accessible, non-rationed information; autistic polymaths remain unexplored.
- Savants can have multiple abilities (e.g. an outstanding musician who is also a strong chess player), but the possibility of autistic polymaths has not been explored, with such multi-domain abilities mentioned only in passing in the literature.
“The idea of an autistic polymath has just not been explored at all.”
- Savant skills cluster in specific areas — music, hyperlexia, three-dimensional drawing, calendar and numerical calculation — because these involve information that tends to be available in most environments and is hard to take away, which is why such abilities develop and persist while others may be rationed out by interventions.
“Because this is information that tends to be available in most environments, and that's often quite hard to take out of those environments.”
Autism research genetics have yielded limited insights despite favorable conditions; personality and autism are largely orthogonal.
- Autism is essentially orthogonal to personality; the literature connecting autism to personality traits is very weak, and any association likely arises from some third factor — at most, personality among autistics may be higher variance.
“Autism is sort of orthogonal to personality. The two are not related.”
- Research on rare and de novo copy variants is a cautionary tale against simplistic explanations of autistic genetics: even in the most favorable setting (the Simons Simplex Collection of single-autistic families) little was found, and the field has seesawed back toward common variants and more complex views where rare and common variants can coexist.
“even looking for them in the most favorable circumstance possible — that is, you're looking in families that only have one autistic person — they did not find much. But that was the leading story for a long time.”