Summary
In 1985, Baron-Cohen, Leslie, and Frith published a study that would dominate autism research for three decades. Using the Sally-Anne test (a false-belief task requiring children to predict behaviour based on false belief rather than reality), they found 80% of autistic children failed, compared to 14% of neurotypical children and 14% with Down syndrome. They concluded autistic children lack a “theory of mind”—the ability to attribute mental states to others.
Baron-Cohen later coined “mindblindness” to describe this deficit: autistic people cannot understand that others have thoughts, beliefs, and feelings. This framed social difficulties as arising from a fundamental cognitive absence—not a processing difference, but a missing capacity.
This framework shaped clinical practice, diagnostic criteria, and public understanding for thirty years. Social skills training assumed autistic people needed to be taught from scratch to understand other minds. The harm is documented; the scientific foundations have eroded.
What the evidence shows
The original finding and its dominance
The 1985 Sally-Anne study was elegant and influential. It provided a clean, testable account of autistic social difficulty: if you cannot represent what someone else believes, you cannot predict their behaviour, and social interaction becomes impossible.
The framework was extended into a developmental model: first-order false belief (understanding that someone holds a belief different from reality), second-order false belief (understanding that someone holds a belief about someone else’s belief), and higher-order mentalising. Autistic people were claimed to be impaired at each level.
Theory of Mind became the dominant paradigm in autism research. It generated hundreds of studies, informed diagnostic practice, and shaped how clinicians, teachers, and parents understood autistic behaviour.
The empirical unravelling
The problems with ToM are not merely theoretical. The empirical foundations have been systematically dismantled.
Not all autistic people fail ToM tasks. In the original study, 20% of autistic children passed; subsequent studies show higher pass rates, especially among autistic adults and those with higher cognitive abilities. A core deficit of autism should be universal. It isn’t.
Gernsbacher and Yergeau (2019) conducted the most comprehensive review of the evidence to date. Their findings are damning. Over 75% of the top 500 Google Scholar articles on “theory of mind” and “autism” simply assert the claim rather than providing original data. Seminal findings have failed to replicate. The various ToM tasks fail to correlate with each other — performance on one task does not predict performance on another supposedly measuring the same construct. ToM scores fail to predict autistic traits, social interaction quality, or empathy. The claim that autistic people lack a theory of mind is, they conclude, “empirically questionable and societally harmful.”
Warnell and Redcay (2019) tested this convergence problem directly by administering diverse ToM tasks to three age groups: preschoolers, school-age children, and adults. Across all ages, the tasks showed minimal correlations with each other. High performance on one ToM measure did not predict performance on another. If ToM were a single underlying capacity, its measures should cohere. They don’t. The implication: “theory of mind” may not be a single construct at all, but a label pasted over a collection of unrelated cognitive processes.
Altschuler and Faja (2021) examined test-retest reliability of cognitive, affective, and spontaneous ToM tasks in autistic school-age children. Reliability was poor — the same child tested twice might pass one day and fail the next. A measure that cannot reliably classify the same individual cannot be the basis for claims about a fundamental cognitive deficit.
Plastow (2012) raised a more philosophical objection: ToM proposes “an ideal of limpid understanding of the other” that neurotypical people don’t achieve either. The false-belief paradigm assesses a second-person phenomenon (understanding another mind) through a third-person method (observing a puppet show). The experimental design doesn’t capture what it claims to measure.
Alternative explanations for task failure are robust. A 2024 systematic review found executive function (cognitive flexibility, working memory) predicts ToM task performance when controlling for language. The Sally-Anne test demands more than mentalising: holding multiple representations, inhibiting knowledge, processing narrative. Autistic children may fail for reasons unrelated to understanding others.
The double empathy reframing shows the difficulty is bidirectional: neurotypical people are equally poor at understanding autistic minds. If ToM were genuinely absent in autistic people, there’d be no reverse pattern. The evidence suggests a mismatch between neurotypes, not a deficit in mentalising.
Predictive processing offers a domain-general explanation: autistic brains weight sensory evidence differently relative to prior predictions. This single mechanism accounts for sensory differences, insistence on sameness, and social processing differences—without needing a broken “social cognition module”.
Autistic pragmatic competence shows they correctly interpret indirect speech acts (“Can you pass the salt?” as a request, not a question about ability). A 2024 study found autistic and non-autistic adults equally use discourse context to determine speaker intent. These findings contradict a framework claiming autistic people fundamentally cannot understand others’ communicative intentions.
The harm
The “mindblindness” framing caused measurable damage. Social skills programmes built on ToM assumptions teach autistic people to perform neurotypical behaviour (eye contact, small talk, reading expressions) on the premise that they lack basic capacity to understand others. The evidence shows short-term compliance but not lasting satisfaction, and contributes to masking.
Autistic people told they are “mindblind” internalise inability to connect with others, contributing to depression, anxiety, and reduced self-advocacy. When autistic social cognition is framed as absent rather than different, genuine strengths (loyalty, directness, pattern-based understanding, deep engagement) go unrecognised.
The ToM framework directed decades of research toward documenting what autistic people cannot do rather than what they do. The double empathy problem, monotropism, and autistic-autistic communication research emerged only after ToM began losing its hold.
What replaced it
No single framework has replaced Theory of Mind, but several contribute to a more accurate picture:
- The double empathy problem — social difficulty as bidirectional mismatch, not one-sided deficit
- Monotropism — attention architecture (fewer interests, greater intensity) explains social differences through an attentional lens rather than a mentalising one (see Autistic social experience)
- Predictive processing — domain-general differences in precision weighting account for both sensory and social processing differences
- The alexithymia hypothesis — much of what’s attributed to “autistic empathy deficit” is actually alexithymia (difficulty identifying one’s own emotions), which co-occurs with autism in ~50% of cases but is a separate construct (see Alexithymia and autism)
Open questions
Is there any genuine mentalising difference in autism, once executive function, language, and alexithymia are controlled for? The evidence is mixed. Some studies find residual differences; others find none. The answer may depend on how “mentalising” is measured and whether the measure assumes neurotypical social processing as the standard.
Key sources
- Baron-Cohen, S., Leslie, A.M., & Frith, U. (1985). “Does the autistic child have a ‘theory of mind’?” Cognition, 21(1), 37–46.
- Baron-Cohen, S. (1995). Mindblindness: An Essay on Autism and Theory of Mind. MIT Press.
- Milton, D. (2012). “On the ontological status of autism: the ‘double empathy problem.‘” Disability & Society, 27(6).
- Beall, P. & Monaghan, P. (2015). “Children with Autism Understand Indirect Speech Acts.” PLOS ONE.
- Applied Neuropsychology: Child (2024). Systematic review on executive function, ToM, and language in autism.
- Gernsbacher, M.A. & Yergeau, M. (2019). Empirical failures of the claim that autistic people lack a theory of mind. Archives of Scientific Psychology, 7(1), 102–118.
- Warnell, K.R. & Redcay, E. (2019). Minimal coherence among varied theory of mind measures in childhood and adulthood. Cognition, 191, 103997.
- Altschuler, M. & Faja, S. (2021). Brief report: Test-retest reliability of cognitive, affective, and spontaneous theory of mind tasks. J Autism Dev Disord.
- Plastow, M. (2012). ‘Theory of mind’ II: Difficulties and critiques. Australasian Psychiatry.