Autism and ASD: Current State of Knowledge and Main Theories Under Discussion

What classifications and research currently establish

A.1 — What clinical classifications group under an ASD diagnosis

In the World Health Organization’s ICD-11, autism spectrum disorders are classified among neurodevelopmental disorders. Diagnostic requirements notably concern persistent difficulties in reciprocal social interaction and communication, together with restricted, repetitive, or inflexible patterns of behaviour, interests, or activities. The classification also distinguishes presentations according to intellectual development and functional language. [S1]

DSM-5-TR likewise uses a single category of Autism Spectrum Disorder with specifiers, rather than several separate diagnoses as in some earlier versions.

Diagnosis is therefore clinical and developmental: it is based on the person’s history, observation of behavioural and developmental characteristics, and their impact. There is currently no single biological test that, by itself, can establish or rule out an ASD diagnosis in an individual. Research involving imaging, EEG, eye-tracking, genetics, transcriptomics, metabolism, immunity, or the microbiome has not produced a universal diagnostic biomarker. [S2–S4]

This point matters: classification among neurodevelopmental disorders is not the same thing as diagnosis through a single neurological marker.

A.2 — The “spectrum” is multidimensional

The word “spectrum” does not describe a simple scale running from “a little” to “a lot.” Two people who both meet ASD criteria may differ greatly in language, adaptive abilities, autonomy, sensory characteristics, interests, motor functioning, health, measured cognition, associated conditions, and support needs.

The literature distinguishes several forms of heterogeneity:

• phenotypic, when observable manifestations differ;
• developmental, when trajectories do not follow the same course;
• etiological, when different genetic or biological architectures are associated with the diagnosis;
• clinical, when associated conditions such as intellectual disability, epilepsy, ADHD, sleep disorders, motor disorders, or language difficulties vary greatly between individuals.

This heterogeneity is now a central fact of research and a major difficulty for any theory claiming to explain all the people grouped under the same diagnosis.

A.3 — Genetics: an important contribution, but distributed and heterogeneous

Family, twin, and genomic studies support an important genetic contribution to the probability of receiving an ASD diagnosis. But this contribution does not correspond to an “autism gene.” [S5–S7]

The literature describes, among other things:

• numerous common variants with very small individual effects;
• rare variants that may have larger effects;
• de novo mutations;
• copy-number variations;
• certain chromosomal abnormalities or monogenic variants;
• combinations of rare variants and polygenic contributions.

The same variant may be associated, depending on the individual, with very different presentations: an ASD diagnosis, intellectual disability, epilepsy, language difficulties, or no major clinical manifestation. Conversely, people with similar behavioural profiles may have different genetic architectures.

Genetic data therefore demonstrate a strong and complex association between many genetic factors and diagnosed populations; they do not provide a single, simple cause.

A.4 — “Syndromic,” “non-syndromic,” “idiopathic,” “secondary”

Part of the literature uses the expression “syndromic autism” when ASD criteria occur in the context of an identified genetic, chromosomal, or molecular syndrome. Frequently studied examples include fragile X syndrome, tuberous sclerosis complex, and several syndromes related to chromosomal or monogenic variations. [S8–S10]

The expression “non-syndromic” is used when no overall syndrome has been identified. It does not mean “non-genetic,” “homogeneous,” or “without biology.”

The term “idiopathic” generally indicates that no specific etiology has been identified. “Primary” and “secondary” are used more variably, and their use is not perfectly standardized. [S11]

Other historical subtypings have also been proposed. The “essential autism / complex autism” distinction proposed by Judith Miles and colleagues relied notably on dysmorphology or microcephaly and did not become a general classification. [S12–S13]

These expressions are therefore tools belonging to particular research traditions or to clinical genetics, not universal divisions of autism.

A.5 — Biology and biomarkers: group differences without a universal individual test

Neuroscience describes many average differences between groups, but these do not converge on a single lesion or signature. Findings vary according to methods, ages, tasks, subgroups, and associated conditions.

A 2026 neurophysiological meta-analysis covering 145 EEG/MEG studies and more than 7,000 participants found several average differences in sensory processing, generally of modest magnitude and with substantial heterogeneity. The authors emphasize the current limitations of using these measures clinically as biomarkers. [S4]

It is therefore useful to distinguish four concepts that are often conflated:

• genetic: DNA variations and their transmission or occurrence;
• biological: a much broader set of bodily phenomena;
• neurological: phenomena concerning the nervous system;
• biomarker: a measure that may help detect, predict, or monitor a state.

An average biological difference is not automatically an individual diagnostic test.

A.6 — Laurent Mottron and the study of perceptual and cognitive particularities

The work of Laurent Mottron and collaborators has a distinctive place in research on autistic cognition. It helped shift some debates away from an exclusively deficit-based reading toward the study of processing modes that may include enhanced performance on certain tasks.

The Enhanced Perceptual Functioning (EPF) model notably proposes greater availability of local processing, greater autonomy of certain perceptual operations, and a substantial contribution of perception to tasks usually regarded as “cognitive.” [S14]

A functional-imaging meta-analysis found, across several visual tasks, relatively greater recruitment of perceptual regions and lower frontal recruitment in autistic groups. [S15]

These findings do not mean that “autistic perception is superior” in every domain. Other tasks may show difficulties or different strategies. The main point is that the profile cannot be reduced to a general impairment.

A.7 — Intelligence, talents, and “veridical mapping”

A 2007 study found that some autistic participants performed markedly better on Raven’s Progressive Matrices than on certain Wechsler scales, leading the authors to discuss the possibility that some instruments may underestimate forms of autistic reasoning. [S16]

Work on savant abilities and veridical mapping subsequently proposed that some talents may rely on detecting or constructing highly precise structural correspondences between sets of information. [S17–S18]

These studies do not constitute a general theory of all autistic people. They propose mechanisms for particular profiles and particular forms of learning.

A.8 — Prototypical autism and developmental bifurcations

Mottron and Gagnon proposed the notion of prototypical autism, referring to a qualitatively recognizable developmental presentation that is narrower than the very broad set currently grouped under an ASD diagnosis. [S19]

Research on “frank presentations” has examined whether experienced clinicians may rapidly recognize some highly typical presentations, while also showing that such an impression cannot replace a full assessment and does not cover all diagnosed people. [S20]

In 2025, Mottron and several collaborators proposed the framework of asymmetric developmental bifurcations, according to which some human variants might correspond to relatively stable alternative developmental trajectories. The authors include the hypothesis that some prototypical forms of autism may fall within this type of bifurcation. [S21]

This framework is recent and does not constitute a consensus.

A.9 — Weak central coherence and processing styles

The theory of weak central coherence, associated notably with Uta Frith and Francesca Happé, was initially formulated as a reduced spontaneous tendency to integrate local information into a global or contextual whole.

Over time, the idea was reformulated in terms of a processing style rather than a global deficit: global processing ability may be present while local information receives a different priority.

This evolution shows that the same experimental observation can be described in different ways depending on the theoretical framework used.

A.10 — Bayesian models and predictive processing

In 2012, Elizabeth Pellicano and David Burr proposed that some perceptual particularities observed in autistic people might be related to lower weighting of certain prior expectations (priors) relative to current sensory information. [S22]

Many related models were subsequently proposed: greater precision assigned to sensory signals, more strongly weighted prediction errors, difficulties estimating contextual volatility, or differences in learning regularities.

A 2023 review of 83 studies found highly mixed results. [S23] A 2026 meta-analysis found a small-to-moderate average effect in the direction predicted by a simple Bayesian model, but with substantial heterogeneity, limiting the idea of a universal mechanism. [S24]

The literature therefore supports investigating possible differences in the relationship between expectations, context, and current signals, but not the simple formula that autistic people “have no priors.”

A.11 — Statistical learning, categorization, and rationality

A recent systematic review of statistical learning found variable results depending on modality, task, and whether learning was explicit or implicit. [S25]

A 2024 meta-analysis of category learning found, on average, lower performance in some experimental paradigms, but with substantial heterogeneity. [S26] These findings do not, by themselves, justify concluding that there is a general inability to “understand categories” or to generalize.

Conversely, several decision-making studies have found reduced susceptibility to some classic biases, such as framing or decoy effects, leading some authors to use the expression “enhanced rationality.” [S27–S29]

This expression refers to results in specific tasks; it does not mean that autistic people are globally more rational or free of biases.

A.12 — Social interaction and the “double empathy problem”

The double empathy problem, proposed by Damian Milton and subsequently developed in several empirical studies, challenges the idea that difficulties in social understanding are solely a deficit property of the autistic person.

Studies indicate that non-autistic people may also have difficulty understanding the intentions, emotions, or communication styles of autistic people. [S30–S31]

The concept has therefore introduced a more relational and bidirectional approach to interaction difficulties. The literature is still developing, however, and does not provide a single mechanism explaining all interactions.

A.13 — Neurodiversity: descriptive concept, paradigm, and movement

Recent historical work emphasizes that the concept of neurodiversity developed collectively within autistic and neurodivergent communities rather than being the isolated invention of one person. [S32]

It is useful to distinguish:

• neurodiversity as the fact of diversity in neurocognitive functioning;
• the neurodiversity paradigm, which includes normative positions;
• the term “neurodivergent,” a broad non-diagnostic social category whose boundaries vary with usage.

Relationships among neurodiversity, disability, medicalization, rights, and identity remain debated. [S33–S35]

A.14 — “Profound autism” and support needs

The 2021 Lancet Commission popularized the expression “profound autism” to draw attention to people requiring very substantial and long-term assistance.

In 2026, a Delphi process proposed a research definition intended to improve comparability across studies. [S36]

This category is mainly functional and descriptive. A genomic, transcriptomic, and regulomic study published in 2026 did not identify a clear biological separation corresponding to the group defined in this way. [S37]

Research on differential diagnosis also emphasizes the need, as far as possible, to distinguish characteristics related to ASD criteria from those related to intellectual disability and from those arising from other conditions or difficulties. [S38]

A.15 — What appears relatively robust, and what remains open

At the date of this publication, several findings can be regarded as relatively robust:

• current ASD diagnoses rely on clinical and developmental criteria;
• populations grouped under those criteria are highly heterogeneous;
• genetic associations are substantial but distributed across very many mechanisms;
• no single biomarker explains or diagnoses the whole group;
• some perceptual, cognitive, developmental, and social differences are reproducible in certain tasks;
• several findings contradict exclusively deficit-based descriptions;
• context and the identity of the interaction partner have become legitimate objects of research on social interaction;
• support needs vary considerably and cannot be inferred from a single measure.

Several fundamental questions nevertheless remain open:

• do people currently grouped under an ASD diagnosis share a common mechanism?
• does the current clinical category bring together several distinct developmental realities?
• what mechanisms connect perception, learning, language, behaviour, and interaction?
• which differences are primary, which are adaptive, and which are related to environment or associated conditions?
• at what level of analysis should the best explanations be sought: genes, cells, neural networks, perception, cognition, development, interaction, environment, or a combination of several levels?

The current documentary state of knowledge is therefore rich in observations and models, but it does not provide a single theory capable of accounting for everything currently studied under the categories “autism” and “ASD.”