Autism — documentary reference points on the current state of knowledge and the main theories discussed

Purpose: to preserve and explain precisely the main scientific and theoretical notions discussed in the September 2026 work.

Scope: clinical framework, heterogeneity, genetics, syndromic and non-syndromic autism, work by Laurent Mottron, perceptual and Bayesian models, categorization, rationality, double empathy, neurodiversity and “profound autism”.

The document systematically distinguishes diagnostic classifications, empirical observations, explanatory models, etiological hypotheses, and philosophical or political positions.

It does not claim to be a general encyclopedia of autism science. References are placed as close as possible to the passages they document and are directly clickable.

Executive summary

This documentary note distinguishes several levels of knowledge that are often mixed together: clinical classifications, empirical findings, cognitive or developmental models, etiological hypotheses, and conceptual or political debates. It does not propose a single theory of autism. Its purpose is to preserve precise reference points on the notions discussed, what they actually establish, and their limitations.

Clinically, ICD-11 and DSM-5-TR group very diverse profiles under an autism diagnosis. Diagnosis remains based on developmental history and clinical observation; there is no single biomarker that can establish autism in an individual person. The “spectrum” is multidimensional: heterogeneity concerns, among other things, language, measured intelligence, autonomy, sensory characteristics, co-occurring conditions, developmental trajectories, support needs and etiological bases. Genetic data show a substantial but extremely heterogeneous contribution, combining common variants, rare variants, de novo mutations, chromosomal variations and, in some situations, identifiable genetic syndromes.

Laurent Mottron's work occupies a particular place in the study of autistic cognition: Enhanced Perceptual Functioning, autistic intelligence, veridical mapping, structured learning, prototypical autism and then asymmetric developmental bifurcations. This work has helped show that some perceptual or cognitive characteristics are not necessarily reducible to deficits, while leaving many questions open. Other traditions study the weight of context, expectations and regularities. Weak central coherence was reformulated from a global deficit toward a processing style; Bayesian models, since Pellicano and Burr, propose different weighting of some prior expectations. The evidence available in 2026 provides partial but highly heterogeneous support and does not justify either the idea that autistic people would not use “priors” or the idea that they would always perceive reality more accurately.

In the social domain, the “double empathy problem” has helped recognize that difficulties in understanding between autistic and non-autistic people can be relational and bidirectional. Neurodiversity has transformed social and ethical frameworks, while “neurodivergent” remains a very broad social category rather than a homogeneous explanatory category. Finally, the debate on “profound autism” reminds us that a category may be useful for identifying very substantial support needs without corresponding to a distinct biological subtype: a Delphi research definition was published in 2026, while a recent genomic, transcriptomic and regulomic analysis found no clear biological separation of the group thus defined.

1. Status, purpose and limits of the document

This document is an independent documentary archive. It presents classifications, research findings, theoretical models and controversies from external institutions, researchers, clinicians and movements. It does not present the personal conceptions developed in the project “Autism — The End of the Puzzle” and does not use them as an interpretive framework.

It does not claim to summarize all autism science. Such a goal would be unrealistic: the literature is immense and covers genetics, development, perception, cognition, language, communication, health, education, environment, interventions, public policy, social sciences and many other fields. This document focuses on the questions discussed here: the scientific and clinical status of autism, heterogeneity, syndromic and non-syndromic autism, idiopathic and secondary autism, genetics and biomarkers, Laurent Mottron's work, Enhanced Perceptual Functioning, veridical mapping, prototypical autism, Bayesian models and predictive processing, categorization, “enhanced rationality”, double empathy, neurodiversity and “profound autism”.

One reading rule is essential: several levels that are often mixed together must constantly be distinguished.

1- A diagnostic classification describes criteria for placing a group of people under the same clinical category.

2- An empirical observation establishes that an average difference has been found in a given task or measure.

3- A theoretical model attempts to explain several observations through a common mechanism.

4- An etiological hypothesis concerns causes or developmental trajectories.

5- An administrative or research category may be useful without constituting a “natural kind” or a distinct biological category.

6- A philosophical, social or political position may be important without being an experimental conclusion.

The fact that a word appears in a scientific publication therefore does not mean that everything implied by that word has been demonstrated.

2. Clinical framework, spectrum, heterogeneity, genetics and etiological categories

2.A. How clinical science currently defines autism

In the World Health Organization's ICD-11, autism is classified among neurodevelopmental disorders. Diagnostic requirements mainly concern persistent differences in the ability to initiate and sustain reciprocal social interaction and communication, as well as restricted, repetitive and inflexible behaviors, interests or activities. ICD-11 also distinguishes presentations according to intellectual development and functional language.

Source: World Health Organization. Clinical descriptions and diagnostic requirements for ICD-11 mental, behavioural and neurodevelopmental disorders (CDDR). 2024. ISBN 978-92-4-007726-3. ↗

DSM-5-TR likewise uses a single category of “autism spectrum disorder” with specifiers rather than the former separate categories such as Asperger syndrome, autistic disorder, etc.

It is important to understand what this means and what it does not mean.

It means that in clinical practice, diagnosis is based on a set of developmental and behavioral characteristics assessed in context.

It does not mean that there is a lesion, a single brain abnormality, a molecule, a gene or a biological test whose presence would be sufficient to establish the diagnosis.

Autism diagnosis therefore remains essentially clinical and developmental. Biomarker research is extensive, but no single biological measure is currently specific, sensitive, validated and reproducible enough to replace clinical assessment. Recent work examines brain imaging, EEG, eye-tracking, genetics, transcriptomic, proteomic, metabolomic and immune profiles, the microbiome and various digital signatures, but these avenues remain at very different levels of validation.

Sources: Understanding autism: Causes, diagnosis, and advancing therapies. 2025. PMID 40449388. ; Wang BM et al. Presymptomatic Biological, Structural, and Functional Diagnostic Biomarkers of Autism Spectrum Disorder. J Neurochem. 2025. PMID 40390287. ; Early autism detection: a review of emerging technologies, biomarkers, and explainable AI approaches. 2025/2026. PMID 41449406. ↗ ↗ ↗

This distinction is important: “classified as neurodevelopmental” is not equivalent to “diagnosed neurologically”.

Sources: Understanding autism: Causes, diagnosis, and advancing therapies. 2025. PMID 40449388. ; Wang BM et al. Presymptomatic Biological, Structural, and Functional Diagnostic Biomarkers of Autism Spectrum Disorder. J Neurochem. 2025. PMID 40390287. ; Early autism detection: a review of emerging technologies, biomarkers, and explainable AI approaches. 2025/2026. PMID 41449406. ↗ ↗ ↗

2.B. The word “spectrum”: what it means and what it does not mean

The word “spectrum” is often understood as if there were a single line running from “slightly autistic” to “very autistic”. That is not a good representation of current clinical use.

The spectrum is multidimensional. Two people may both meet diagnostic criteria while differing greatly in language, measured intelligence, autonomy, sensory characteristics, interests, communication, adaptive abilities, co-occurring disorders, physical health, developmental trajectory and support needs.

This great heterogeneity is now one of the central problems in research. It complicates the search for common mechanisms and can produce contradictory findings when studies use very different groups while calling them by the same diagnosis.

Source: Mottron L, Bzdok D. Autism spectrum heterogeneity: fact or artifact? Mol Psychiatry. 2020. PMID 32123308.

Several kinds of heterogeneity must be distinguished.

First, phenotypic heterogeneity: people can differ markedly in observable characteristics.

Second, developmental heterogeneity: trajectories are different.

Some children show recognizable characteristics very early; in others they become clear later; some show regression of certain abilities, others do not.

Third, etiological heterogeneity: multiple genetic, biological and developmental combinations may be associated with autistic presentations.

Fourth, heterogeneity of co-occurring conditions: intellectual disability, epilepsy, sleep disorders, motor disorders, language disorders, anxiety, ADHD and other characteristics may coexist with autism but are not present in everyone.

It is therefore risky to take a characteristic observed in one subgroup and turn it into a definition of all autism.

2.D. Genetics: what is well established

The genetic contribution to autism is strong. Family, twin and genomic studies converge on this point. Contemporary reviews often give heritability estimates in the range of 70 to 90%, depending on methods and populations.

Source: Genovese A, Butler MG. Genetic contributions to autism spectrum disorder. Psychol Med. 2020/2021. PMID 33634770. ↗

However, “strongly genetic” does not mean “determined by a single gene”.

The genetic architecture of autism is extraordinarily complex. It includes:

— very many common variants, each with a very small individual effect, whose effects partly add up as polygenic risk;

— rare variants that may have a larger effect;

— de novo mutations, meaning mutations that arose in the child without being detected in the parents;

— copy-number variations of chromosomal segments;

— chromosomal abnormalities;

— monogenic variants in some situations;

— combinations of rare variants and polygenic risk.

Sources: Genovese A, Butler MG. Genetic contributions to autism spectrum disorder. Psychol Med. 2020/2021. PMID 33634770. ; Trost B et al. The contributions of rare inherited and polygenic risk to ASD in multiplex families. 2023. PMID 37506195. ; Molecular and Genetic Mechanisms in Autism Spectrum Disorder. Ann Neurol. 2025. PMID 40801227. ↗ ↗ ↗

More than one hundred risk genes have been strongly implicated through rare variants, and the number varies with criteria and the progress of research. But most population-level risk does not come from a few rare mutations: inherited common variants also play a major role.

Sources: Genovese A, Butler MG. Genetic contributions to autism spectrum disorder. Psychol Med. 2020/2021. PMID 33634770. ; Trost B et al. The contributions of rare inherited and polygenic risk to ASD in multiplex families. 2023. PMID 37506195. ; Molecular and Genetic Mechanisms in Autism Spectrum Disorder. Ann Neurol. 2025. PMID 40801227. ↗ ↗ ↗

There is therefore no single “autism gene”.

The same genetic variant can also produce very different outcomes across people: autism, intellectual disability, epilepsy, language disorder, a mild phenotype, or even no major clinical manifestation. This is the issue of incomplete penetrance and variable expressivity.

Conversely, two people with fairly similar autistic presentations may have very different genetic architectures.

2.E. “biological”, “genetic”, “neurological” and “biomarker” are not synonyms

This distinction is fundamental when reading articles.

“Genetic” concerns DNA, genetic variants and their transmission or emergence.

“Biological” is much broader: it may include genetics, cellular development, gene expression, physiology, immunity, metabolism, the nervous system, etc.

“Neurological” specifically concerns the nervous system.

A “biomarker” is a biological measure that may be associated with a state or help detect, predict or monitor it.

A phenomenon may have genetic or biological bases without there being a single diagnostic biomarker. Likewise, the existence of average brain differences between groups does not mean that one can examine the brain of an individual person and reliably decide whether that person is autistic.

The neurophysiological meta-analysis published in 2026 illustrates this well: across 145 EEG/MEG studies and more than 7,000 participants, some sensory-response latencies differed on average between autistic and non-autistic groups, but effects were generally modest, heterogeneity was substantial and clinical application as a biomarker remained limited.

Source: Ghosh A et al. Neurophysiological alterations during sensory processing in autism — a meta-analysis. Eur Child Adolesc Psychiatry. 2026;35:767–784. DOI 10.1007/s00787-025-02917-0. ↗

2.F. “Syndromic” autism

The term “syndromic autism” belongs mainly to clinical genetics and developmental neurology. It generally refers to an autistic presentation occurring in the context of an identifiable genetic, chromosomal or molecular syndrome that also includes other characteristics.

Frequently cited examples include fragile X syndrome, tuberous sclerosis complex, certain chromosomal variations, some microdeletions or microduplications, and several monogenic syndromes.

Sources: Fernandez BA, Scherer SW. Syndromic autism spectrum disorders: moving from a clinically defined to a molecularly defined approach. Dialogues Clin Neurosci. 2017/2018. PMID 29398931. ; Moss J, Howlin P. Autism spectrum disorders in genetic syndromes: implications for diagnosis, intervention and understanding the wider ASD population. J Intellect Disabil Res. 2009. PMID 19708861. ; Heterogeneity of Autism Characteristics in Genetic Syndromes: Key Considerations for Assessment and Support. 2023. PMID 37193200. ↗ ↗ ↗

Historically, a syndrome was first recognized through a clinical constellation: particular morphology, involvement of several organs, developmental profile, etc., then confirmed by a targeted test. Modern sequencing techniques introduced a second situation: some people have a pathogenic molecular variant discovered through genomic analysis without the clinical picture having previously been recognized as a classic syndrome. One review therefore proposed distinguishing “clinically defined” from “molecularly defined” syndromes.

Source: Fernandez BA, Scherer SW. Syndromic autism spectrum disorders: moving from a clinically defined to a molecularly defined approach. Dialogues Clin Neurosci. 2017/2018. PMID 29398931. ↗

The term “syndromic autism” is not itself a distinct ICD-11 diagnosis.

Sources: Moss J, Howlin P. Autism spectrum disorders in genetic syndromes: implications for diagnosis, intervention and understanding the wider ASD population. J Intellect Disabil Res. 2009. PMID 19708861. ; Heterogeneity of Autism Characteristics in Genetic Syndromes: Key Considerations for Assessment and Support. 2023. PMID 37193200. ↗ ↗

2.G. “Non-syndromic” autism

“Non-syndromic” generally means that no identifiable overall syndrome has been recognized.

Above all, it does not mean:

— absence of genetics;

— absence of biology;

— absence of familial factors;

— “pure” autism in a philosophical sense;

— a homogeneous category.

The non-syndromic group itself probably contains many different genetic and developmental architectures.

The boundary between syndromic and non-syndromic also shifts with advances in genetic analysis. A person formerly described as “idiopathic” may later receive a more precise genetic explanation.

2.H. “idiopathic”, “primary” and “secondary”

The term “idiopathic” essentially means that no specific identified cause is available.

In some texts, “primary autism” is used in a similar way for autism that is not explained by another identified condition.

“Secondary autism” is used when the autistic presentation occurs in the context of a recognized cause or condition: genetic syndrome, metabolic disease, neurological injury or another situation considered etiologically relevant.

Source: Casanova MF et al. Editorial: Secondary vs. Idiopathic Autism. Front Psychiatry. 2020. PMID 32346372. ↗

These terms are not perfectly standardized and can be confusing.

The main problem with the word “secondary” is that it can suggest a simple causal chain: “condition X → autism”. Yet the actual mechanisms may be much more complex. In some genetic syndromes, behaviors resembling some autism criteria are observed, but the profile may differ qualitatively from non-syndromic autism. The effects of intellectual disability, language, epilepsy or the syndrome-specific phenotype must then be distinguished. [R10-R11, R13]

2.I. “essential autism” and “complex autism”

Judith Miles and colleagues proposed in the early 2000s a distinction between “essential autism” and “complex autism”.

In their definition, the “complex” group notably showed signs of early morphogenesis abnormalities, particularly significant dysmorphology and/or microcephaly. The remainder was classified as “essential”. The initial study reported differences in prognosis, epilepsy, MRI and familial recurrence.

Source: Miles JH et al. Essential versus complex autism: definition of fundamental prognostic subtypes. Am J Med Genet A. 2005. PMID 15887228. ↗

This classification did not become the general diagnostic system for autism.

Later studies reproduced some differences but also showed the limitations of the dysmorphology tool and its dependence on the populations studied.

Source: Identification of Essential, Equivocal and Complex Autism by the Autism Dysmorphology Measure: An Observational Study. 2020. PMID 32767173. ↗

It should therefore be regarded as a historical attempt at subtyping rather than a universally accepted division.

2.J. A difficult question: can different causes produce the same “autism”?

Yes, at the level of the current diagnosis, this is possible.

The criteria are phenotypic: they describe developmental and behavioral characteristics. Several biological pathways can therefore lead to behaviors that meet the same criteria.

This creates a classic problem in medicine and psychiatry: a clinical category may be useful without corresponding to a single cause.

This is one reason why some researchers want to stratify cohorts further: by genetic etiology, intellectual development, language, trajectory, perceptual characteristics, comorbidities, etc.

But this stratification creates another problem in turn: the more finely groups are divided, the greater the risk of losing possible common mechanisms.

3. Laurent Mottron, perception and prototypical models

3.A. Laurent Mottron: general place in autism research

Laurent Mottron is a professor and researcher in autism psychiatry and cognition in Montreal. Much of his work has challenged exclusively deficit-based interpretations of autistic functioning and studied perceptual, cognitive and developmental characteristics as forms of organization that can include enhanced performance.

His work does not constitute “the official theory of autism”. It occupies an important but specific place in a field where many models coexist.

Several periods or lines of work can be distinguished:

— Enhanced Perceptual Functioning;

— autistic intelligence and reasoning;

— savantism and veridical mapping;

— learning through structures;

— criticism of spectrum heterogeneity;

— prototypical autism;

— asymmetric developmental bifurcations.

3.B. Enhanced Perceptual Functioning: the EPF model

The “Enhanced Perceptual Functioning” (EPF) model, presented notably in the 2006 synthesis by Mottron, Dawson, Soulières, Hubert and Burack, partly reverses models that explained autistic perception mainly through a deficit in global processing.

Source: Mottron L, Dawson M, Soulières I, Hubert B, Burack J. Enhanced perceptual functioning in autism: an update, and eight principles of autistic perception. J Autism Dev Disord. 2006;36:27–43. PMID 16453071. DOI 10.1007/s10803-005-0040-7. ↗

It proposes several principles, including:

— a local orientation more frequently available by default;

— enhanced discrimination for some simple visual or auditory stimuli;

— greater autonomy of perceptual operations from higher-level processing;

— a greater contribution of perception to some cognitive tasks considered complex;

— greater recruitment of perceptual regions in some tasks.

It should be noted that “local” does not automatically mean “incapable of global processing”. An important idea in the model is that local perceptual processing may be preferential or more autonomous without global processing necessarily being absent.

A 2012 functional-imaging meta-analysis found, across several visual tasks, greater recruitment of occipital, temporal and parietal regions in autistic participants and relatively less frontal recruitment. This finding is compatible with a greater role for perceptual mechanisms in some tasks.

Source: Samson F, Mottron L, Soulières I, Zeffiro TA. Enhanced visual functioning in autism: an ALE meta-analysis. Hum Brain Mapp. 2012;33:1553–1581. PMID 21465627. ↗

But the perceptual literature is not uniform. For example, a meta-analysis of global motion found a small average disadvantage in some coherent or biological-motion tasks.

Source: Van der Hallen R et al. Global Motion Perception in Autism Spectrum Disorder: A Meta-Analysis. 2019. PMID 31489542. ↗

The reasonable conclusion is therefore not “autistic perception is superior”, but rather: some perceptual domains show reproducible differences, including advantages, disadvantages or different strategies depending on the task.

3.C. Autistic intelligence and reasoning tests

Mottron, Michelle Dawson and colleagues have also studied how intelligence tests assess autistic people.

In a well-known 2007 study, autistic participants scored markedly higher on Raven's Progressive Matrices than on Wechsler scales, particularly among some participants classified as having low intelligence by Wechsler. The authors argued that instruments constructed from non-autistic profiles could underestimate some forms of autistic reasoning.

Source: Dawson M et al. The level and nature of autistic intelligence. Psychol Sci. 2007. (Raven/Wechsler study). ↗

This study does not demonstrate that Raven represents “true intelligence” or that all autistic people are underestimated. It shows that profiles across tests can be atypical and that interpretation of a global score should be cautious.

3.D. Savantism and “veridical mapping”

Mottron, Dawson and Soulières proposed that some exceptional abilities observed in autistic people may be related to the greater role of perception and a strong ability to detect structures.

Source: Mottron L, Dawson M, Soulières I. Enhanced perception in savant syndrome: patterns, structure and creativity. Philos Trans R Soc Lond B. 2009. PMID 19528021. ↗

The notion of “veridical mapping”, developed more specifically in 2013, refers to mapping isomorphic structures: stable relations between elements can be detected across different systems, for example between written symbols and sounds, musical pitches and note names, calendars and numerical regularities.

Source: Mottron L et al. Veridical mapping in the development of exceptional autistic abilities. Neurosci Biobehav Rev. 2013;37:209–228. PMID 23219745. ↗

The model has been used to account for phenomena such as:

— hyperlexia;

— absolute pitch;

— calendrical calculation;

— some savant abilities;

— spontaneous learning based on structured regularities.

Its theoretical interest lies in viewing these abilities as positive consequences of detection and mapping mechanisms rather than as isolated curiosities.

However, veridical mapping does not automatically apply to all autistic people and has not been demonstrated as a single mechanism of autism.

3.E. Language and structured information in Mottron's work

In more recent work, Mottron, Ostrolenk and Gagnon proposed that some autistic children may develop components of language not only through social exposure to spoken language, but from highly structured information such as letters, numbers, songs, subtitles, grapheme-phoneme correspondences or other regular systems.

Source: Mottron L, Ostrolenk A, Gagnon D. In Prototypical Autism, the Genetic Ability to Learn Language Is Triggered by Structured Information, Not Only by Exposure to Oral Language. Genes. 2021;12:1112. DOI 10.3390/genes12081112. ↗

This proposal is linked to observations of hyperlexia and atypical language trajectories.

It is a developmental model emphasizing the existence of learning pathways different from the socially mediated trajectory usually described.

3.F. Prototypical autism according to Mottron and Gagnon

In 2023, Mottron and Gagnon proposed a more radical reformulation: “prototypical autism”.

Source: Mottron L, Gagnon D. Prototypical autism: New diagnostic criteria and asymmetrical bifurcation model. Acta Psychol. 2023;237:103938. PMID 37187094. ↗

Their starting point is that the progressive broadening of criteria has produced a highly heterogeneous “spectrum” category. They nevertheless argue that there is a particularly recognizable clinical presentation in early childhood, with qualitatively specific signs.

They propose that this prototypical autism corresponds to an asymmetric developmental bifurcation.

In their model, some core elements would be organized around a reduction in the usual social bias and increased investment in complex or structured non-social information.

The model differs from a simple theory of “severity”. Prototypical autism is not synonymous with “severe” autism.

It also differs from the idea of a single genetic cause. The authors argue that a category may be recognizable by its developmental configuration without having a single molecular etiology.

This model is not incorporated into ICD-11 or DSM-5-TR. It is a controversial scientific proposal.

3.G. “Frank autism”: can a prototype be recognized quickly?

Before and alongside Mottron's work, researchers studied the notion of “frank autism”, meaning the very rapid clinical impression that a person presents an obvious autistic picture.

A survey of expert clinicians found that many reported forming such an impression quickly. But a rapid clinical impression can be subject to confirmation bias, so it must be tested and never substituted for a full assessment.

Source: De Marchena A, Miller J. ‘Frank’ presentations as a novel research construct and element of diagnostic decision-making in autism spectrum disorder. 2017. PMCID PMC5400744. ↗

A study published in 2024 asked clinicians to make impressions after five minutes of observation. The impressions often agreed with the full diagnostic assessment and correlated with observed severity, but sensitivity was insufficient: a substantial number of autistic people were not recognized as such in the first minutes.

Source: Canale RR et al. Investigating frank autism: clinician initial impressions and autism characteristics. Mol Autism. 2024;15:48. PMID 39538274. ↗

This finding supports the existence of a subgroup whose presentation is very immediately recognizable, but it does not demonstrate that this subgroup alone constitutes “true autism”. It also shows why a clinical prototype can have good specificity and limited sensitivity.

3.H. Critique of the continuum: “autism-ness does not exist, but autism does”

In 2025, Mottron, Lafrance McGuire and Gagnon published an explicit critique of the idea that autism is simply the end of a quantitative continuum of traits distributed throughout the population.

Source: Mottron L, Lafrance McGuire O, Gagnon D. Autism-ness Does Not Exist, but Autism Does. Part 1. Autism & Developmental Language Impairments. 2025;10. PMID 41356733. ↗

They propose distinguishing:

— isolated traits that may exist to varying degrees in many people;

— a qualitatively recognizable prototypical autistic configuration.

They describe prototypical autism as a quasi-categorical possibility, stable in human development and evolution.

This position conflicts with a substantial part of contemporary dimensional research. It raises difficult questions:

— where to place the boundary;

— how to address people whose presentation is less typical;

— how to avoid biases related to sex, gender, masking, age or intelligence;

— how to validate a prototype independently of expert intuition;

— how to avoid turning a heuristic tool into arbitrary diagnostic exclusion.

3.I. The 2025 “asymmetric developmental bifurcations”

In September 2025, Mottron and several collaborators proposed a broader framework: “asymmetric developmental bifurcations” or ADB.

Source: Mottron L et al. Asymmetric developmental bifurcations in polarized environments: a new class of human variants, which may include autism. Mol Psychiatry. 2025;30:6155–6164. PMID 41023420. DOI 10.1038/s41380-025-03275-8. ↗

The model draws on dynamic systems in which the same system can reach two different stable states depending on its development.

The authors compare prototypical autism with some relatively stable minority human variants, such as left-handedness or certain developmental variants, and propose that some forms of autism may be understood as an alternative developmental organization rather than as a progressive disease or a simple accumulation of deficits.

This proposal is very recent. It is intellectually important because it shifts autism toward a theory of developmental variants, but it still requires broad testing. It is not a consensus.

4. Perception, prediction, learning, categorization and rationality

4.A. Weak central coherence: another theoretical tradition

Before Mottron's models, Uta Frith and Francesca Happé developed the theory of “weak central coherence”.

In its earlier form, it suggested that autistic people would have a weaker tendency to spontaneously integrate local information into a global or contextual meaning.

The theory was later reformulated to speak more of a cognitive style or processing bias than of an obligatory deficit: stronger local processing may provide advantages in some tasks, while global-processing ability may be present but less spontaneously prioritized.

This historical evolution is instructive: the same experimental observation can first be interpreted as a lack of integration, then as a preference for or autonomy of local processing.

Mottron's EPF model partly arose in opposition to the deficit interpretation of weak central coherence, even though the two traditions sometimes describe closely related experimental phenomena.

4.B. Pellicano and Burr: “when the world becomes too real”

In 2012, Elizabeth Pellicano and David Burr applied Bayesian models of perception to autism.

Source: Pellicano E, Burr D. When the world becomes ‘too real’: a Bayesian explanation of autistic perception. Trends Cogn Sci. 2012;16:504–510. PMID 22959875. ↗

In a simplified Bayesian framework, perception depends on combining two sources:

— current sensory evidence;

— “priors”, meaning expectations built from previous experience.

Pellicano and Burr proposed that priors might be weighted less strongly in autistic people. If sensory input is less corrected or pulled toward prior expectations, perception may sometimes remain closer to what is currently present.

They used a formulation that became famous: the world might become “too real”, because perception would be less modulated by what is expected.

This idea has been very influential, but it was originally a theoretical hypothesis rather than a general demonstration.

4.C. Predictive coding and other Bayesian models

After 2012, several variants of the model were proposed.

Some emphasize not consistently weaker priors, but the precision assigned to sensory signals.

Others propose prediction errors that are weighted too strongly.

Still others emphasize difficulty estimating contextual volatility: if the world seems difficult to predict, the system may place less confidence in acquired regularities.

These models are related but not identical. It is therefore incorrect to speak of a single “Bayesian theory of autism”.

4.D. What does the Bayesian evidence say in 2026?

A comprehensive review published in 2023 examined 83 studies conducted during the ten years following Pellicano and Burr's article. Results were highly mixed: a slight majority found no difference in the integration of priors, and studies were often underpowered or methodologically heterogeneous. Differences appeared more often for priors learned during the experiment than for previously established knowledge.

Source: Angeletos Chrysaitis N, Seriès P. 10 years of Bayesian theories of autism: A comprehensive review. Neurosci Biobehav Rev. 2023;145:105022. PMID 36581168. ↗

A meta-analysis published in 2026 synthesized 23 studies and found a small-to-moderate average effect in the direction predicted by the “simple Bayesian model” (Hedges g ≈ 0.37), but with substantial heterogeneity not explained by the moderators examined. The authors conclude that support for a simple universal model is limited and recommend more sophisticated hierarchical models.

Source: Cui K et al. A Systematic Review and Meta-analysis of Empirical Evidence for the Simple Bayesian Model of Autism. Neuropsychol Rev. 2026;36:184–195. PMID 40576893. DOI 10.1007/s11065-025-09672-8. ↗

A neuroimaging meta-analysis published in September 2026 also examines prediction and prediction-error networks in autism. Its existence confirms that this question remains actively studied; it does not thereby turn predictive-coding hypotheses into a single explanation of autism.

Source: Prediction and Prediction Error in Autism: A Meta-Analysis of Functional Magnetic Resonance Imaging Results. Biological Psychiatry Global Open Science. September 2026;6(5):100760. DOI 10.1016/j.bpsgos.2026.100760. ↗

The current assessment is therefore nuanced:

— there is evidence compatible with different weighting between expectations and sensory signals;

— the effect is neither universal nor homogeneous enough to summarize all autism;

— learning regularities and the precision of predictions may matter more than the simplistic idea of “weak priors”.

4.E. Statistical learning and regularities

Statistical learning is the ability to extract probabilistic regularities from the environment.

A systematic review published at the end of 2025 re-examined this field in autism. Again, the findings show a complex picture: depending on task type, modality, and whether learning is explicit or implicit, differences are not uniform.

Source: Bell RR et al. A systematic review of statistical learning in autism spectrum disorder. Molecular Autism. 2026;17:2. Published 24 December 2025. ↗

This is important because a person may have an excellent ability to detect some structured regularities while using some contextual or social probabilities differently.

4.F. Categorization and category learning

A 2024 meta-analysis pooled 50 studies, with more than 1,200 autistic participants and 1,400 non-autistic participants. On average, it found lower performance among autistic participants in experimental category-learning tasks, with a medium effect size and substantial heterogeneity.

Source: Wimmer L et al. Category learning in autistic individuals: A meta-analysis. Psychon Bull Rev. 2024;31:460–483. PMID 37673843. ↗

This result should not be overinterpreted.

A “category learning task” is a specific experimental paradigm. It may depend on abstraction of a prototype, explicit rules, probabilities or other strategies.

The result therefore does not, by itself, justify saying that autistic people “cannot generalize” or “do not understand categories”.

Rather, it raises more specific questions:

— what information is used to form a category;

— how much within-category variation is tolerated;

— how much weight is given to the prototype compared with individual exemplars;

— which strategies are favored;

— under what conditions a generalization is considered relevant.

4.G. “Enhanced rationality”

Several decision-making studies have produced a result opposite to exclusively deficit-based descriptions: some classic decision biases appear less strongly in autistic participants.

In 2008, De Martino and colleagues showed lower sensitivity to some framing effects.

Source: De Martino B et al. Explaining enhanced logical consistency during decision making in autism. J Neurosci. 2008. PMID 18923049. ↗

In 2017, Farmer and colleagues showed greater consistency of choices under some manipulations involving decoy alternatives (“decoy effect”).

Source: Farmer GD et al. People With Autism Spectrum Conditions Make More Consistent Decisions. Psychol Sci. 2017. PMID 28635378. ↗

A 2021 review grouped several findings under the term “enhanced rationality”.

Source: Rozenkrantz L et al. Enhanced rationality in autism spectrum disorder. Trends Cogn Sci. 2021. PMID 34226128. ↗

The expression should be used cautiously. It does not mean that autistic people are globally more rational, more logical or free from bias.

It means that within some specific tasks, several contextual biases that usually affect decision-making may have a weaker effect.

Whether this results from weaker contextual influence, more deliberative processing, an emotional difference, a different perception of the options or other mechanisms remains debated.

4.H. Sensory perception: neither simple “hypersensitivity” nor simple superiority

Sensory characteristics are now incorporated into diagnostic criteria. They include hyperreactivity, hyporeactivity and sensory seeking.

The scientific literature shows average differences across several domains, but there is no single autistic sensory profile.

The 2026 EEG/MEG meta-analysis already mentioned included 145 studies. It found longer latencies for some early components but no robust overall amplitude differences, with substantial heterogeneity.

Source: Ghosh A et al. Neurophysiological alterations during sensory processing in autism — a meta-analysis. Eur Child Adolesc Psychiatry. 2026;35:767–784. DOI 10.1007/s00787-025-02917-0. ↗

These data are incompatible with two opposite simplifications:

— “the autistic person simply receives too much information”;

— “the autistic person simply has superior senses”.

The phenomena appear to depend on stimulus type, processing level, response timing, language, age, co-occurring conditions and strategies.

5. Social communication and double empathy

5.A. The “double empathy problem”

Damian Milton proposed the “double empathy problem” in 2012.

The theory challenges the idea that difficulties in social understanding between autistic and non-autistic people are entirely caused by an autistic deficit in empathy or theory of mind.

It proposes that some misunderstandings are relational and bidirectional: two people with very different experiences, conventions and communication styles may have reciprocal difficulty understanding one another.

This proposal stimulated experimental research. Studies show that non-autistic people may also be less accurate in inferring the states or intentions of autistic people, or may evaluate autistic behaviors negatively.

Sources: Mitchell P, Sheppard E, Cassidy S. Autism and the double empathy problem: Implications for development and mental health. Br J Dev Psychol. 2021. PMID 33393101. ; Cheang RTS et al. Do you feel me? Autism, empathic accuracy and the double empathy problem. Autism. 2024. PMID 38757626. ↗ ↗

A 2024 study of empathic accuracy produced evidence compatible with this approach.

Source: Cheang RTS et al. Do you feel me? Autism, empathic accuracy and the double empathy problem. Autism. 2024. PMID 38757626. ↗

The literature is still developing, however. The double empathy problem does not yet have a single mechanistic theory capable of predicting every interaction. Some reviews emphasize that the concept is promising but still insufficiently defined to explain all social difficulties by itself.

Source: Autism — the problem of double empathy. Review. PMID 41696836. ↗

Source: Mitchell P, Sheppard E, Cassidy S. Autism and the double empathy problem: Implications for development and mental health. Br J Dev Psychol. 2021. PMID 33393101. ↗

6. Neurodiversity and disability

6.A. Neurodiversity: origins

The neurodiversity movement and concept took shape in neurodivergent self-advocacy communities, particularly autistic communities, during the 1990s.

They were long attributed mainly to Judy Singer, but recent historical work has shown a more collective genesis. A 2024 article emphasizes that the notions of “neurological diversity” and later “neurodiversity” were developed collectively by several people from these communities.

Source: Botha M et al. The neurodiversity concept was developed collectively: An overdue correction on the origins of neurodiversity theory. Autism. 2024;28:1591–1594. PMID 38470140. ↗

The word subsequently spread far beyond autistic circles and into universities, companies, human-resources policies, education, mental health, autism services and public debate.

6.B. Three meanings not to confuse: neurodiversity, paradigm and neurodivergence

First meaning: neurodiversity as a descriptive fact. A human population displays diversity in neurocognitive functioning.

Second meaning: the neurodiversity paradigm. It adds a normative position: a neurological or cognitive difference should not automatically be equated with a pathology to be eliminated; the person's abilities, environment, quality of life, preferences and rights should be examined.

Third meaning: “neurodivergent”. This term is used for a person whose functioning is regarded as different from the so-called neurotypical norm.

The boundaries of this third meaning are much less stable. Depending on authors and communities, “neurodivergent” may include autism, ADHD, dyslexia, dyspraxia, dyscalculia, Tourette syndrome, and sometimes psychiatric, acquired or much broader conditions.

This extension makes the term useful as a social coalition but much less precise as a scientific category.

6.C. Neurodiversity and disability

Historically, the neurodiversity movement is not simply a denial of disability. A substantial proportion of its authors and activists also identify with the social or relational model of disability and the disability rights movement.

However, the relationship between neurodiversity and disability remains contested.

Several positions coexist:

— some people regard their difference as a disability;

— others say they are disabled mainly by the environment;

— others personally reject the identity of a disabled person;

— others emphasize the combination of a valued difference and real limitations;

— some families or people with very substantial needs criticize discourses they regard as too centered on forms of autism compatible with high autonomy.

The 2024 humanities literature specifically highlights the political issues involved in using neurological language as a basis for identity, rights or access to resources.

Source: Jones EK, Orchard V. Neurodiversity and disability: what is at stake? Med Humanit. 2024;50:456–465. PMID 38360797. ↗

6.D. Critiques of the “neuro-”

Several authors have criticized the way the prefix “neuro-” can appear more objective than it really is.

One criticism is neurocentrism: explaining complex human phenomena mainly through the brain can erase social, developmental, cultural and relational dimensions.

Another concerns epistemic injustice: the people directly concerned may be excluded from defining their own experience; conversely, some identity-based discourses can also become norms that delegitimize people with different experiences of the same condition.

Gauld, Jurek and Fourneret therefore proposed in 2024 thinking more broadly in terms of “biopsychosocial diversity”, in order to avoid both overmedicalization and undermedicalization.

Source: Gauld C, Jurek L, Fourneret P. Diversity, epistemic injustice and medicalization. Cortex. 2024;176:234–236. PMID 38580533. ↗

Other authors have analyzed how the “neuro” can be co-opted both by neuroscience and by identity movements.

Source: Co-opting the “neuro” in neurodiversity and the complexities of epistemic injustice. Cortex. 2023. PMID 37837731. ↗

7. “Profound autism”, support needs and clinical differentiations

7.A. “Profound autism”: where does the term come from?

The 2021 Lancet Commission proposed the administrative or descriptive term “profound autism” to draw greater attention in research and public policy to autistic people requiring very substantial and long-term assistance.

The initial problem was that the category did not have a uniformly applied definition.

Studies could use different IQ thresholds, different language criteria or different autonomy criteria, making comparisons difficult.

7.B. The 2026 research definition

In June 2026, Siegel, Lord, Tager-Flusberg and colleagues published the results of a Delphi process aimed at establishing a common research definition.

Source: Siegel M et al. Developing a consensus research definition for profound autism using a modified Delphi method. Mol Autism. 2026;17:28. PMID 42380932. DOI 10.1186/s13229-026-00727-y. ↗

The proposed definition requires:

— an autism diagnosis;

— an age of at least 8 years;

— adaptive functioning far below the expected level, with inability to perform most activities of daily living independently;

— adult supervision required to preserve health, safety and well-being;

— and either severely impaired cognitive abilities, with an IQ below 50, or verbal communication limited to single words or fixed expressions used mainly for basic needs, or both.

76% of participants in the final round approved the definition.

The authors themselves emphasize important limitations: participants were predominantly American, access to assessment tools is unequal, and worldwide application is difficult.

It is a research definition, not a new universal WHO diagnostic category.

7.C. A needs category is not necessarily a biological category

A study published in August 2026 analyzed genomic, transcriptomic and regulomic data to determine whether people classified as having “profound autism” formed a distinct molecular group.

Source: Genomic, Transcriptomic, and Regulomic Analyses Do Not Support Profound Autism as a Distinct Biological Category. 2026. PMID 42282515. ↗

The authors found no evidence of a clear biological separation corresponding to this category.

They did, however, observe some differences in regulatory networks associated in particular with being speaking or non-speaking.

This result must be interpreted precisely.

It does not prove that “profound autism” is useless.

Nor does it prove that very substantial needs do not exist.

It indicates that the proposed clinical or functional boundary does not correspond, in these data, to a clear genetic or molecular boundary.

A category can therefore be useful for research on support needs without constituting a biological subspecies of autism.

7.D. Autism, intellectual disability and differential diagnosis

An important difficulty is that some autism criteria describe the absence or atypical acquisition of expected developmental abilities.

In a person with substantial intellectual disability, some social and communication difficulties may also arise for general developmental reasons.

The diagnosis of autism therefore theoretically requires that social and behavioral characteristics are not explained solely by the person's general developmental level.

The literature emphasizes that this distinction can be very difficult, particularly in genetic syndromes and in people with little spoken language.

Source: Thurm A et al. State of the Field: Differentiating Intellectual Disability From Autism Spectrum Disorder. Front Psychiatry. 2019. PMID 31417436. ↗

This difficulty probably contributes to the heterogeneity of research cohorts.

8. Methodological precautions and state of knowledge

8.A. What group studies can and cannot tell us

Most research compares groups.

If an autistic group obtains an average result different from that of a non-autistic group, this does not mean that every person in the first group has that characteristic.

Two distributions can overlap extensively while still having statistically different means.

This is particularly important for concepts such as:

— sensory sensitivity;

— central coherence;

— Bayesian priors;

— rationality;

— intelligence;

— categorization;

— empathy.

A serious theory of autism must therefore avoid turning an average difference into a universal individual essence.

8.B. Association, correlation and causation

Another common pitfall is to confuse three levels.

Association: a characteristic occurs more often with autism.

Correlation: two measures vary together.

Causation: one characteristic directly produces the other.

For example, finding a genetic variant more frequently in autistic people does not necessarily mean that this variant “causes autism” in a simple and deterministic way.

Likewise, observing a brain difference in autistic adults does not always make it possible to know whether that difference is an early cause, a consequence of development, an adaptation, the effect of different experiences, or a mixture of these factors.

8.C. Why theories of autism often contradict one another

Several reasons explain the apparent contradictions.

The groups studied are not the same.

The tasks do not measure exactly the same thing.

Children and adults may show different phenomena.

People with and without intellectual disability are sometimes mixed together or, conversely, excluded.

Historically, samples have included many more boys and men.

Diagnostic criteria have changed over time.

Researchers use different theoretical models to interpret the same result.

Negative studies are sometimes less likely to be published.

Sample sizes are often small.

Some theories are formulated at a very general level but tested using very narrow tasks.

Research from 2020–2026 increasingly tends to acknowledge these problems of heterogeneity and reproducibility explicitly.

8.D. What is relatively robust in 2026

Without claiming absolute consensus, several propositions are now relatively robust.

— Autism is developmental and emerges early in the life trajectory, even if identification may occur late.

— Current diagnosis is clinical and behavioral, not based on a single biomarker.

— Genetic contributions are substantial but extremely heterogeneous.

— There are forms associated with identifiable genetic syndromes or variants.

— Sensory, cognitive, language and adaptive profiles are highly variable.

— Some perceptual tasks show superior performance in some autistic groups; others show difficulties.

— Differences in the processing of context, categorization, prediction or decision-making are real in some tasks, but they do not yet form a universal theory.

— Difficulties in social interaction can no longer be studied solely as isolated properties of the autistic person; the role of the interaction partner and context is now a legitimate object of research.

— Support needs vary enormously and cannot be inferred from a single measure.

8.E. Important but non-definitive models

Enhanced Perceptual Functioning: an influential model, supported by several perceptual and imaging findings, but not universal.

Veridical mapping: a fruitful proposal for some forms of learning and some abilities, but not a general theory.

Weak central coherence: a historical model progressively reformulated as a cognitive style; useful but insufficient on its own.

Bayesian / predictive processing: an important family of models, with partial and highly heterogeneous empirical support.

Enhanced rationality: a set of interesting findings in some decision-making tasks, but not a general superiority.

Double empathy problem: a major relational correction to the one-way social-deficit model, with growing empirical evidence but mechanisms still under discussion.

Prototypical autism: a recent and structured proposal, but not validated as a new consensual clinical category.

Asymmetric developmental bifurcation: a recent theory of a developmental variant; highly speculative at this stage.

8.F. Scientific questions that remain open

Several fundamental questions remain.

Is there one common mechanism, or are there several common mechanisms, across all people currently diagnosed as autistic?

Does the current diagnostic category group together several different developmental realities?

Can reproducible subgroups be identified without creating arbitrary boundaries?

What proportion of perceptual differences arises from the sensory level itself, and what proportion from contextual weighting, attention, learning or experience?

How do perceptual characteristics translate into differences in language, behavior and social interaction?

Are differences in priors global or specific to certain domains and certain types of learning?

Which characteristics are primary, and which result from adaptation to the environment?

How can people with little spoken language be studied scientifically without confusing absence of response, absence of understanding, and motor or communication difficulties?

What is the best level of analysis: genes, neural networks, perception, cognition, development, interactions, or a combination of these levels?

8.G. An important precaution concerning the word “explanation”

None of the theories presented in this document should be described as “the explanation of autism”.

A theory may explain one phenomenon without explaining autism as a whole.

EPF explains some perceptual findings.

Predictive processing attempts to explain some relationships between perception, expectations and uncertainty.

Double empathy explains part of relational difficulties.

Genetics explains a substantial part of risk variation and some identifiable etiologies.

“Profound autism” organizes functional support needs.

These levels are not interchangeable.

9. Terminological reference points and conclusion

9.A. Quick guide to terms

Syndromic autism: autism associated with an identifiable genetic, chromosomal or molecular syndrome; a term used in research and clinical genetics.

Non-syndromic autism: no overall syndromic cause identified; does not mean “non-genetic”.

Idiopathic autism: no specific etiology identified.

Secondary autism: a variable term referring to an autistic presentation associated with a recognized condition or cause; terminology is not perfectly standardized.

Essential autism / complex autism: historical subtyping by Miles based notably on dysmorphology and microcephaly; not official.

Enhanced Perceptual Functioning: Mottron model emphasizing the autonomy and sometimes increased efficiency of certain perceptual processes.

Veridical mapping: model of exact structural correspondences involved in some forms of learning and exceptional abilities.

Weak central coherence: theory of a lower spontaneous priority given to global/contextual integration; later reformulated as a processing style.

Hypo-priors: hypothesis that some prior expectations have a lower weight in autistic perception.

Predictive processing: family of models in which the brain generates predictions and processes errors between predictions and incoming signals.

Enhanced rationality: weaker expression of some classic biases in certain decision-making tasks.

Double empathy problem: model according to which part of autistic/non-autistic social difficulties is bidirectional and relational.

Prototypical autism: Mottron and Gagnon's proposal for a qualitatively recognizable developmental core; unofficial and controversial.

Frank autism: rapid clinical impression of a highly recognizable autistic presentation; an empirically studied phenomenon, but insufficient for detecting all cases.

Asymmetric developmental bifurcation: recent theory by Mottron and colleagues treating certain human variants, potentially including prototypical autism, as alternative developmental states.

Neurodiversity: diversity of neurocognitive functioning; can also refer to a paradigm and a social movement.

Neurodivergent: broad, non-diagnostic social term with variable boundaries.

Profound autism: descriptive/research category aimed at autistic people with very substantial support needs; Delphi research definition published in 2026; not a universal official category.

9.B. Documentary conclusion

The current state of knowledge does not provide a single theory of autism.

Instead, it provides several layers of knowledge.

Clinical science can identify, with reasonable but imperfect reliability, a set of trajectories and characteristics grouped under the category “autism”.

Genetics establishes a major contribution and a highly heterogeneous architecture, including common variants, rare variants and identifiable syndromes.

Neuroscience detects average differences in the structure, function and timing of processing, but no single marker summarizes autism.

Cognitive sciences identify reproducible differences in some domains: perception, contextual processing, categorization, prediction, decision-making, learning and communication.

Mottron's models have played an important role in showing that some perceptual and cognitive differences are not necessarily deficits and can support enhanced abilities.

Bayesian models have provided a powerful way to formalize the relationship between sensory signals and expectations, but the evidence available in 2026 does not support a simple and universal model.

The double empathy problem shifted part of social research from the question “what is missing in the autistic person?” toward “what happens between two different modes of communication and experience?”.

Neurodiversity has profoundly changed ethical and social frameworks, while introducing new conceptual ambiguities when it becomes a very broad meta-category.

Finally, debates on prototypical autism and profound autism illustrate two opposing but complementary movements: some researchers seek to narrow the category around a developmental prototype, while others seek to distinguish people with extreme needs so that they do not disappear within the heterogeneity of the spectrum.

None of these approaches has settled the question “what is autism?”.

In September 2026, the most accurate scientific answer therefore remains pluralistic: autism is a robust but highly heterogeneous developmental clinical category, associated with complex genetic and biological bases, particular perceptual and cognitive profiles, and several partial explanatory models, none of which yet accounts for the whole.