Theories of autism

In 1985 Simon Baron-Cohen, Alan Leslie and Uta Frith proposed that autistic children lack a "theory of mind", the ability to think about what other people believe and know. They tested it with a false-belief story, now known as the Sally-Anne task, in which a doll's marble is moved while she is out of the room.¹ Of the 20 autistic children tested, 16 (80%) said the doll would look where the marble was now, so about one in five passed.^1,3 Baron-Cohen's 1995 book Mindblindness proposed that autistic children are "mindblind" as a result of a selective impairment in mindreading.²

Group differences on these tasks are well replicated. A 2019 meta-analysis of 75 studies of adults (3,361 autistic adults, about three quarters of them male, and 5,344 non-autistic adults) found a large difference on theory-of-mind tasks, pooled from 39 of those studies.⁶ But the difference is neither universal nor specific to autism. Passing tracks verbal ability: in one pooled analysis, typical children had an even chance of passing two false-belief tasks at a verbal mental age of about 4, autistic participants only at about 9.⁴ A 1998 paper reporting three meta-analyses found that people with intellectual disability, as well as autistic people, score below typically developing people on these tasks, so the difficulty is not unique to autism; how large the differences were depended on the comparison group, such as whether its intellectual disability came from Down syndrome, and on how groups were matched.⁵ A 2026 systematic review and network meta-analysis of 89 studies (9,038 participants across 11 psychiatric conditions) found lower scores on theory-of-mind tasks that use static illustrations in every condition except the relatives of people with bipolar disorder; the difference was moderate in autism and largest in schizophrenia, which scored significantly below autism, and it varied with the kind of mentalizing task. The review calls the prominent ceiling effects of these tasks a major limitation.⁴⁰ Because many autistic people without intellectual disability passed false-belief tasks, researchers designed harder "advanced" tests; Gnanathusharan Rajendran and Peter Mitchell note that autistic groups do less well on these but do not fail outright.³

The core claim is now disputed. In a 2019 review, Morton Ann Gernsbacher and the autistic scholar M. Remi Yergeau argue that the claim that autistic people lack a theory of mind is empirically questionable, because theory-of-mind tasks often fail to agree with each other and key findings have not replicated.⁷

Uta Frith's weak central coherence theory, first set out in 1989, proposed that autistic people have difficulty pulling details together into an overall meaning.³ In 2006 Francesca Happé and Frith revised it: the detail focus may be a bias rather than a deficit, and it may sit alongside social differences without explaining them.⁸ Enhanced perceptual functioning, from Laurent Mottron and colleagues, holds that autistic perception is more locally oriented, with stronger low-level perception that runs more independently of higher-level thinking.⁹

The largest test of both ideas, a 2015 meta-analysis of 56 articles with about 1,000 autistic participants, found neither enhanced local visual processing nor a deficit in global processing. What differed was speed: autistic participants reached the global level more slowly.¹⁰ The detailed evidence, including puzzles, pitch and intelligence tests, is in Cognitive styles.

Executive functions are the skills used to plan, switch between tasks and hold back a response. In 1991 Sally Ozonoff, Bruce Pennington and Sally Rogers reported widespread executive and harder (second-order) theory-of-mind difficulties in autistic people without intellectual disability compared with a matched clinical group, while difficulties with simpler theory-of-mind tasks appeared only in some of them.¹¹ By 2004 Elisabeth Hill described whether executive dysfunction is primary in autism as "a topic of much debate".¹²

A 2018 meta-analysis of 235 studies (14,081 participants, 6,816 of them autistic) found a moderate overall difference, similar across executive domains and fairly stable from childhood to adulthood. The difference was smaller in more recent studies, which the authors suggest may reflect wider diagnostic criteria; separate executive subdomains were not supported, and few measures were sensitive enough to help with diagnosis.¹³ Rajendran and Mitchell's 2007 review argues that the difference is neither universal nor unique. Of the few studies reporting individual rates, one found executive problems in 50% of autistic participants and another found 96% below the comparison average, a liberal criterion. People with ADHD, OCD, Tourette syndrome or schizophrenia perform similarly on some executive tests, though some studies find different profiles.³ The full evidence is in Executive function.

In 2002 Simon Baron-Cohen proposed the extreme male brain theory: autism as an extreme of a "male" profile in which systemizing, the drive to find the rules of a system, outweighs empathizing.¹⁴ On questionnaires the pattern holds on average. In an online survey of 671,606 people, 36,648 of whom reported an autism diagnosis, autistic people of both sexes were shifted on average toward the male-typical profile (more systemizing relative to empathizing), the usual sex differences were smaller among autistic people, and the result held in a second, smaller online sample (226 autistic people) analyzed by the same team. Diagnosis was self-reported, the sample came from a television channel's website, and every measure was self-report.¹⁵ A 2026 meta-analysis of 34 studies found the empathy questionnaire separates autistic and non-autistic groups more than the systemizing questionnaire, and sex differences were smaller among autistic people. Proportional differences between autistic and non-autistic groups were larger among women than among men, and the authors judge that the theory has "accrued extensive support". Its pool includes the survey above.¹⁶

The biological side is weaker. In a Danish birth cohort of 128 autistic boys and matched controls, studied by Baron-Cohen's group, a combined measure of fetal steroid hormones in amniotic fluid was modestly raised; the measure covered several steroids, not testosterone alone.¹⁷ A paper from another research group reported two studies, one of 81 children with a genetic condition that raises prenatal testosterone (and 72 relatives) and one of amniotic testosterone in 92 typically developing preschoolers. These measured parent-reported autistic traits rather than diagnoses, and neither found a significant link.¹⁸ A 2020 systematic review of 22 amniotic testosterone studies found that studies by researchers affiliated with Baron-Cohen were more likely to report fully supportive results, and that none of the conclusions was supported by enough studies, while allowing, as preliminary, that fetal testosterone may play a role in autism-related traits.¹⁹ In brain scans from a large shared dataset, a "brain maleness" score based on the shape of deep (subcortical) brain structures was slightly higher in autism, but the difference disappeared once overall brain size was taken into account.²⁰ The ethicists Timothy Krahn and Andrew Fenton argue that gendering these abilities is unpersuasive and may bias diagnosis toward boys, which they suggest could cost autistic girls and women.²¹

In 2007 Henry Markram, Tania Rinaldi and Kamila Markram proposed what they then called the intense world syndrome, from work with rats exposed before birth to valproic acid, a drug linked to autism when taken in pregnancy. They proposed that local brain circuits are hyper-reactive and hyper-plastic, producing heightened perception, attention and memory, so that the world becomes painfully intense and the person withdraws into routines.²² In 2010 Kamila and Henry Markram restated it as the intense world theory, a unifying theory of the neurobiology of autism, with predictions and a hypothesis that would prove it false.²³

The animal model itself has limits. A 2015 review noted that, like any disease model, the valproic acid model cannot reproduce every feature of autism, though it argued that the model reproduces many of them, and weighed the evidence for and against its validity.²⁴ The theory's picture of an overwhelming sensory world is close to how many autistic people describe sensory life, which is covered in Sensory processing and prediction.

A family of theories treats perception as prediction: the brain combines what it expects with what the senses report. In 2012 Elizabeth Pellicano and David Burr proposed that weaker prior expectations may lead autistic people to perceive the world "more accurately rather than modulated by prior experience".²⁵ Rebecca Lawson, Geraint Rees and Karl Friston proposed in 2014 that autism involves too much weight on sensory evidence relative to expectations.²⁶ Sander Van de Cruys and colleagues proposed the same year that autistic people give prediction errors uniformly high, inflexible weight, and read insistence on sameness as a way of keeping the world predictable.²⁷

Tests have been mixed. A 2021 systematic review of 47 studies found differences in learning which events predict others (especially when the predictive cue was weak or inconsistent), in habituation and in adaptation, and called for narrowing the hypothesis, which it found too broad.²⁸ A 2023 review of ten years of Bayesian theories found the results highly mixed, with a slight majority of studies finding no difference in how autistic people use prior expectations, though reduced influence was reported more often for expectations learned during the experiment than for ones brought in, and it noted low statistical power.²⁹ A 2026 meta-analysis of the fMRI studies that tested prediction found that about three-fifths of the experiments reported no reliable group difference. Of the rest, only the contrasts where non-autistic participants were more active than autistic ones came together in one region, tied to error monitoring and uncertainty. The authors read the pattern as broadly consistent with altered prediction in autism, at high levels of processing, and urge caution given how few contrasts could be pooled and how many null results there were.⁴¹ A 2025 systematic review of statistical, implicit and procedural learning (picking up which events tend to follow which) reached a similar mix: on behavioral tasks autistic participants learned about as well as non-autistic participants apart from slower reactions, while the brain-imaging studies disagreed about the mechanisms; the review notes that its studies rarely include minimally verbal people or people with intellectual disabilities.⁴² The research is covered in Sensory processing and prediction.

Two widely used theories came from autistic scholars. Monotropism was set out in 2005 by Dinah Murray, Mike Lesser and Wenn Lawson (then publishing as Wendy Lawson). It holds that autistic attention gathers into a few interests at a time, and that this is central to autism; the paper drew on autistic people's own accounts, including one of its authors'.³⁰ In a 2023 preprint, not yet peer reviewed, 1,110 people completed a questionnaire written with autistic adults, and both autism and ADHD went with higher scores.³¹ In a 2025 laboratory study of 18 autistic and 22 non-autistic adolescents, the autistic teenagers and their caregivers reported more hyperfocus, but brain-wave measures of hyperfocus did not differ between the groups, and the different measures were mostly unrelated to one another.³⁹

In 2012 Damian Milton proposed the double empathy problem: difficulties between autistic and non-autistic people are a two-way problem of mutual understanding, not a deficit located in the autistic person.³² In a 2020 study of 72 adults passing a story along chains of eight people, autistic chains passed on information as well as non-autistic chains, while mixed chains lost more and reported lower rapport.³³ A larger registered replication by the same group, with 311 adults, published in 2025, did not reproduce the loss in mixed chains: information passed equally well in same-neurotype and mixed chains. Non-autistic chains reported higher rapport, and telling partners about a diagnosis slightly improved rapport, significantly only for the person receiving the information; the authors suggest their more varied sample and changed methods may explain the difference from 2020.³⁴ In 2025 Lucy Livingston and colleagues argued that the idea is loosely defined and cautioned against carrying it into applied settings, such as clinical practice, for now.³⁵

Each of the older theories was first offered as an account of autism as a whole, and none has held up as one. In 2006 Francesca Happé, Angelica Ronald and Robert Plomin argued that it was "time to give up on a single explanation for autism", pointing to twin data that suggested social, communication and repetitive traits are influenced by largely different genes. Those data measured autistic-like traits in a general-population twin sample, not diagnosed autistic people.³⁶ In 2007 Rajendran and Mitchell judged theory of mind, executive dysfunction and weak central coherence against three tests: whether a difference is found in all autistic people, whether it is found only in autism, and whether it explains the condition. They concluded that no single theory integrates all the strands, and favored an account with several separate differences.³ In 2020 Happé and Frith described a shift from autism as one thing to many "autisms", and toward neurodiversity and research done with autistic people.³⁷ A 2026 preregistered review of taxometric studies (tests of whether a condition is a category or a continuum) found that ADHD and language impairment look dimensional, while most autism studies reported a categorical pattern. But those autism studies mostly combined samples from different recruitment sources or selection criteria, such as diagnosed and undiagnosed participants in one sample, a design the review's simulations showed can produce a false categorical result; it treats the question as unresolved and asks for population-based studies.⁴³

Autistic writers have criticized the theories and how they were built. Pete Wharmby retells the Sally-Anne test and calls it "a great example of the kind of huge, broad conclusions that autism researchers (who've historically been overwhelmingly non-autistic) have been prone to make based on the loosest evidence."^c1 Yergeau, a rhetoric scholar, writes that "Mindblindness, then, functions as a rhetorical foil that renders the autistic non-rhetorical at worst, and residually rhetorical at best", and notes the irony of writing a book while "presumably unaware of my reader and my (non)self".^c2

Several writers reframe theory of mind as a two-way gap. Luke Beardon calls it "cross-neurological theory of mind", a close relative of Milton's double empathy, arguing that "it's not valid to suggest that people with autism lack a theory of mind" because the lack runs across neurotypes.^c4 In a handbook chapter, the researchers Charlotte Brownlow, Lindsay O'Dell and Ding Abawi argue that mainstream theories "share a common assumption that there is a deficit in autistic people", and quote a 1998 piece by Perkons contrasting two kinds of theory of mind: "Autistic Theory of Mind = Everyone thinks differently from me".³⁸

Others ask what a theory of autism should do. The autism scholar Nick Chown, in a book that evaluates the theories one by one, proposes three tests of his own: a theory must apply to all autistic people ("if some factor does not apply to all autistic people it has no business being considered as autism theory"), it must describe autism accurately, and it must allow for each person's uniqueness.^c5 Chown concludes that despite "no hard evidence proving any one of the three 'mainstream' theories", each is thought to explain some aspects of autism.^c6 Nick Walker places the deficit theories inside what Walker calls the pathology paradigm, whose root is "the assumption that there is one 'right' style of human neurocognitive functioning".^c7

Tito Mukhopadhyay, a nonspeaking autistic writer who prefers to write by hand and dislikes being called a "typer or typist", recalls meeting many autism experts and sorts researchers into three kinds: those who question, those who observe, and "The ones who will put your head inside machines and pretend to understand the graphs."^c3