Autism-ness Does Not Exist,but Autism Does. Part 1: A Critic of the “Spectrum” Position Used to Describe,Diagnose,and Research Autism,and Its Alternative

Abstract

This article presents the historical roots of the dimensional perspective on autism, the epistemological and clinical critics of its assumptions and effects, and offers an alternative to it. Autism is increasingly being described as the “extreme far end” of a spectrum of traits distributed continuously and heterogeneously throughout the general population, and various comorbid neurodevelopmental conditions. This dimensional perspective, initially a response to the excesses of nominalism in the DSM, creates its own heuristic and clinical dead ends. In contrast with this dimensional paradigm, clinical experts recognize and diagnose prototypical autism based on the high similarity of specific clinical signs that are present during the preschool period. We propose viewing autism as a universal and evolutionarily stable, quasi-categorical possibility of human development, offering a prototypical presentation within a certain age range. We argue that prototypical autism needs to be further clinically described and scientifically investigated before anticipating the inclusion of nonprototypical presentations in an informative “autism spectrum.” To achieve this, instruments based on qualitatively defined signs, with weighted diagnostic value, and universally associated with clinical certainty, must be developed. In the meantime, we recommend that all clinicians suspend the use of DSM-5 clinical specifiers to focus on clinical certainty and the application of differential diagnoses, rather than on the diagnostic thresholds of DSM-5 and of standardized instruments.

Background

The theoretical and clinical development that led from the recognition of autism as a distinct entity, distinguishable from other developmental atypicalities, to its inclusion in/identification to a spectrum in DSM 5, is considered progress by most of the academic community (Pearson et al., 2021). The history of autism, which now spans nearly a century, is often narrated as a shift from a position of uncritical acceptance that autism is a natural category to the gradual discovery of its spectral, heterogeneous, and dimensional nature (Happe & Frith, 2020). This transition “filled” the gap between autism and other developmental variants, as well as between autism and the typical development of a multitude of supposedly intermediate conditions (Mottron & Bzdok, 2020). This is generally presented as progress toward greater truth regarding the nature of autism (Koi, 2021), as a necessary condition for the societal acceptance of neurodiversity (Silberman, 2015) and opening a medically endorsed, legitimate path for people to benefit from accessible services. There are also subjective benefits of identifying with a condition that is socially valued and scientifically endorsed.

The transformation of autism from a category to a spectrum has resulted in a unique increase in reported prevalence. Autism is one of the very few conditions whose reported prevalence has increased by a factor close to 100, from 4 per 10,000 in early textbooks to nearly 4% in the United States, Center of Diseases Control announcements. Its impressive rise in numbers and public consciousness echoes the contemporary questioning about the categorical nature of psychiatric clinical entities, well beyond autism (Haslam et al., 2000).

Most of the characteristics of autism that were accepted as empirical clinical facts half a century ago have been challenged, consistent with, and contributing vastly, to this increase in prevalence. Over the past 20 years, our clinical knowledge of autism, and even our ability to recognize it, has gradually lost the stable benchmarks we had at the end of the 20th century. Almost everything we knew, or thought we knew, about autism is now being questioned: diagnoses are commonly accepted in adulthood without manifestations in childhood (Rodgaard et al., 2021b), the sex ratio in favor of boys is being questioned, autism is condensed into traits that are present in the general population, atypical language development is no longer mandatory for diagnosis, autism no longer even needs to be confirmed by visible manifestations. Furthermore, autistic “traits” are reported in all psychiatric disorders, neurodevelopmental conditions, and learning disabilities, familial and de novo forms are grouped together on the same spectrum, and the subgroups that had been established to bring order to all this have been eliminated. Both the scope and understanding of the concept of autism continues to expand, and to drift further away from the initial clinical descriptions that gave birth to it.

We therefore think it legitimate to question this evolution, which, we fear, could lead to a loss of the autism signal over the next few decades. This article, which radically bucks this historical trend, attempts to break this active or passive unanimity. We will attempt to show that, far from being clinical or historical progress, this constant broadening of social and scientific understanding of autism prevents us from understanding the kind of human variant it constitutes. Such an evolution misleads us about the mechanisms that lead to autism, and is not a necessary condition for the societal progress that concern neurodivergent minorities. We will argue that the apparent consistency between the broadening of diagnostic criteria, the proliferation of self-diagnoses, the fragmentation of the autistic spectrum into traits, and the multiplicity of alleged neurodevelopmental causes are grounded on a paralogism, aggravated by a lack of exposure to a population enriched in autistic individuals. We will address in turn: the dimensional representations of autism, the paralogisms they imply, and our proposed paradigm shift toward prototypical autism in clinical practice and science.

Dimensional Representations of Autism

Sources of the Dimensional Approach

Early in the history of autism, nosography had to contend with the existence of clinical pictures that shared similarities with Kanner's initial description, but also differed from it in a number of ways. This clinical evidence raised the question of heterogeneity, which the dimensional approach promised to address. Domains of heterogeneity can thus be referred to as dimensions. It was probably the story of Asperger's syndrome that shattered the initial categorical evidence and led to a movement that culminated in DSM 5 (Mottron, 2021a). Once the deep connection (which we do not question) between the cases described by Asperger and those described by Kanner was revealed, Lorna Wing, who introduced Asperger's work to the English-speaking world, chose to integrate them into a single category. To do so, she had to substantially alter the description of Asperger's syndrome, bringing it closer to what was then known about autism. Instead of treating the “autism” and the “Asperger” presentations as two related but distinct manifestations of the same variation in human development, she took the highly consequential decision to combine them into a single category.

This opened the door to envision a continuum between autism with profound language delay or atypia, and autism with early language development without any alteration other than in the domain of pragmatics, through clinical presentations increasingly distant from Kanner's autism. Lorna Wing's work foreshadowed the positioning of language as a dimensional specifier rather than a classifier in the DSM-5, which is what we recommend. It also sanctioned a certain level of heterogeneity, establishing heterogeneity as an intrinsic property of autism.

Another justification of dimensional approaches to autism is the notion of autistic traits. One of us has provided an in-depth critique of this in Mottron and Bzdok (2020) and Mottron (2021b). The idea that autistic traits are distributed independently of psychiatric and neurological nosography is closely associated with tools that divide the autistic spectrum into individual questions, each of which is scored and then totaled to create a summary score. This score, identified with severity or “autistic-ness,” can be used to determine a threshold for a categorical diagnosis. Most representative of this trend is the work of Constantino (2011) and his tool for measuring autistic traits, the Social Responsiveness Scale (SRS). This instrument has given rise to countless studies on the presence of “autistic traits” in autistic individuals, relatives, in psychopathology as a whole, but also in the average population. Constantino's position is summarized in his aphorism “autistic traits are continuously distributed in the general population.” We consider this to be the most extreme position among the continuity hypotheses. Constantino's theory is grounded on a vast and deep knowledge of the familial risk for autism, the interpretation of which, however, one can disagree (Mottron et al., 2025). It has evolved to become closer to our discontinuous position recently under the influence of a reconsideration of twin studies (Constantino, 2021).

A decade before Constantino's seminal paper, Simon Baron-Cohen had already put forward a theory and built an empirical tool that produces continuous variables out of clinical reality. “Empathizing” and “systematizing” scales were both the consequence and the justification for his continuous model. Just as Constantino had constructed a single continuous variable, socialization, parameterized by the score of his instrument, Baron-Cohen used two continuous variables, each associated with an instrument that measures them, and promoted these variables as objects of science. In both cases, the living entity to be described and understood is replaced by constructs. Constantino and Baron-Cohen replace a form of life that may be categorical in nature with an abstract entity, a theory-laden proxy, which has only the limits of the extension of its definition, and whose components are also defined in an abstract manner. SBC's and Constantino's theories and ad-hoc instruments equate autism with one impaired (socialization) or one impaired, (empathizing) and one enhanced (systematizing) dimensions. These dimensions are conceptualized independently of autism developmental trajectory, obliterate the clinical discontinuity of prototypicality between at-risk individuals and prototypical cases, attribute properties of their instrument to the condition they measure, and disregard the recognition of a phenotype in favor of its definition, in their terms.

As we will develop in the second section of this article, it is not empathizing or social reciprocity that is a natural category, but autism. Constantino affirms that his instrument's measures of socialization are measures of the degree of autism. Baron-Cohen does the same: by summarizing the complexity of autism through an unequal relationship between two constructs, he replaces the elephant with elephant-ness, or autism with autism-ness. This epistemological sleight of hand has immense consequences. The authority vested in Constantino's assumptions position them as foundations of the current approach to autism in clinical practice, science, and society.

These assumptions may also have contributed to the concept of clinical specifiers in the DSM-5 nosology, such as language, comorbidity, intelligence, and severity. A specifier is a dimension whose variation neither alters nor informs the categorical diagnosis. In the case of language profiles, this means that only the language characteristics that make it possible to score a particular DSM sign (e.g., repetitive behavior) will be relevant to the diagnosis, but that its quantitative variation and developmental course will not inform the diagnosis. The essential content of the nosographic category of autism would thus be preserved even in the presence of variations in one or more of the four specifiers. In other words, one could be equally autistic with a more or less developed level of language, more or less impaired intelligence, with or without the presence of other associated conditions, and finally with a greater or lesser impact on societal adaptation. This is, actually, true, but leads to dramatic mistakes when applied uncritically.

The Nihilistic Position of Waterhouse and Gillberg

The most extreme scientific position questioning the categorical account of autism contests its existence as a distinct entity. This position has been taken by Gillberg (2010) and (Waterhouse & Gillberg, 2014). According to these authors, autism, both as a concept and as a clinical entity, encompasses heterogeneous conditions associated with excessively variable characteristics. Grouping them under a single term would be entirely “cultural” or “subjective.” In consequence, the clinical category “autism,” based neither on objective similarities nor on a shared etiology, would not be the appropriate level of conceptualization in the hierarchical taxonomy of atypical development. It would therefore wrongly lead to the search for common mechanisms and support mode.

According to Gillberg and Waterhouse, it would have been preferable to move up a level in the taxonomic hierarchy of neurodevelopmental atypicalities, from “autism” to the broader “essence” category. This can be contested, as demonstrated by an analogy. The elephant species has no more or less legitimacy than the class of mammals. Both are “true” and describe a certain level of the organization of living beings. However, the class of mammals does not negate the existence of the elephant, whereas Gillberg's “essence” class negates the relevance of autism as a recognizable category. We recognize that the distinction between the logic that prevails in species taxonomy and that of psychiatric conditions is that we do not have to demonstrate that elephants resemble each other, whereas the clinical resemblance leading to diagnostic certainty for autism is suspected of being individual and subjective (Lombardo, 2021). Nevertheless, in the case of elephant, objective and subjective resemblance may converge. Although we will never say or think that autism is another species, since we consider autism to be a human possibility (Mottron et al., 2025), we argue that the epistemic problems associated with its delineation are of the same nature as those raised by the recognition and definition of a species or an organ.

Critique of Dimensional Approaches to Autism

The Implications of the Concept of Dimension

Both the categorical and the dimensional approaches attempt to resolve the problem of heterogeneity. The DSM-5 criteria (and to some extent the standardized diagnostic tools) converge on categorical boundaries based on several distinct thresholds, one per construct (social reciprocity, and repetitive behaviors and restricted interests-RIRB), while determining four dimensions whose variation is irrelevant to the categorical diagnosis, the specifiers. The DSM adds a quantitative polythetic system, where several combinations of the same number of signs allow the threshold for the RIRB area to be reached. The DSM-5 cannot therefore be considered a pure representative of the dimensional approach, as the SRS would be, for example. For the sake of clarity, we will separate the traditionally opposing dimensional and categorical approaches in order to articulate their respective implications.

A categorical approach to autism is based on an identical recognition of a condition despite some variation. It reduces the different presentations of the condition to the idiosyncratic individual variations of same category, whether these variations are quantitative (signs that are more or less pronounced, or numerous) or qualitative (signs that are observed or not, or vary in their presentation). As a result, people with different signs or different levels of these signs could be given the same diagnosis of autism, as long as they have enough of them together. The threshold at which one is confirmed autistic or ceases to be autistic is determined by a number of signs that are considered scarse in the general population. It therefore implies that the distribution of these signs in the general population is not linear. Autistic people exhibit many of these signs, while the rest of the population exhibits few. The distribution of the number of signs would therefore be bimodal, or discontinuous, reflecting the existence of a natural category.

A dimensional approach involves an intermediate construct between the observed condition and the signs that describe/quantify it. Individual signs are reduced to one dimension (e.g., empathizing, systematizing, social reciprocity, RIRB). There are one or two (in the case of DSM-5), three (in the case of DSM-4) or 6 (2 + 4 in the DSM 5, if we include clinical specifiers) dimensions. The score achieved on these dimensions varies continuously according to the number of these signs, which generally have equal diagnostic weight. A threshold will be set by agreement between judges. To prevent too much variability in this system, it will be compartmentalized by establishing minimum thresholds in each of the chosen dimensions. This system is what has been chosen in the DSM, where there must be at least two repetitive signs, and in the ADI, where there must be a minimum number of signs belonging to a list considered to be more discriminating.

In this dimensional approach, the threshold is considered, if not arbitrary, at least a convention, allowing for better interjudge agreement with clinical judgment. A dimensional approach differs from a categorical approach not in terms of the existence or nonexistence of a threshold, but in terms of how it is structured, the value given to it, and above all, the real or conventional nature attributed to it. Above all, unlike the categorical approach, it does not postulate any essential difference between people above or below the threshold, but rather a simple difference in degree.

The Dimensional Approach Involves an Intermediate, Arbitrary Construct Between Unitary Signs and the Diagnosis of Autism

The modeling of autism through the continuous variation of one dimension (e.g., socialization) or a limited number of dimensions, reduces a complex biological object, implying variation in multiple domains, to the continuous value of a single, ad-hoc construct. This construct tends to replace its object in a general movement in which science replaces clinical exposure by collected databases, which loses its power to preselect models and theories compatible with reality. In other words, what applies successfully only to quantifiable traits such as size, weight, or activity level, or in psychology, personality or temperament, may not be suitable for a form of life, inherently multidimensional, even ontologically beyond the notion of dimension. While nobody would identify an elephant to an extreme level of “elephantness,” it seems acceptable to describe somebody as “very autistic” or “slightly autistic.” That is not to say that measuring a continuous dimension cannot, in certain contexts, help to establish conventional “boundaries” when the measured object is inherently quantitative. It does not, however, accurately inform about the occurrence, nature, structure, functioning, or etiology of a complex and stable biological form such as species—and, even more, on its limits. Measuring blood pressure, to which Constantino (2021) uses as a comparison for social functioning in autism, fails to explain the nature and operations related to hemodynamics, in the same extent that measuring social “level” fails to inform on what glue together in the same individual and at the same moment signs as recognizable as hand leading, lateral gazes, tiptoe walking, and lack of response to name.

A descriptive, dimensional reduction may, in some situation, allow the identification of a form of life: the elephant being the largest terrestrial living animal on earth, size may allow to accurately identify elephants in certain situation, though it cannot otherwise make them known to us. You could not build an elephant from the knowledge of its size, without some kind of description. No current construct reliably describes anatomical, behavioral structures or developmental patterns (composed of specific constituents that frequently occur together) allowing for the recognition of autism with clinical certainty. The cost of condensing complex biological forms into quantifiable variables is a loss of the qualities of the observed whole. This cost culminates in the use of “autistic trait” scales to characterize research populations. Autism becomes dimensional only when it is reduced to, and measured by, scales of quantitative traits, condensed into a limited number of quantitative variables, and in fine, in the construction of “autistic-ness.” In autism, the reduction to one-dimensional constructs also obliterate the relationship between the different phenotypic and developmental elements that make up the condition, as well as distorting each of these elements, such as empathy when measured by unidimensional instruments (Cusson et al., 2025). Alternatively, we advocate for the identification of autism as a unique, complex organization of qualitatively discernible signs. We advocate for a reconsideration of the assumption that, in the science of the living word, quantity has some kind of higher scientific value, would be more falsifiable that quality, that would inherently be “subjective.” Quantification can be the laziest way to approach the living world.

Accounting for the differences and commonalities between two biological forms, whether between a caterpillar and a butterfly, or between an autistic individual and a neurotypical person, would require a near-infinite number of dimensions, leading to the aptly termed curse of dimensionality. The choice among this infinite array of dimensions (and consequently, their weighting in a diagnostic algorithm), is arbitrary and depends on the population used to validate these algorithms. Categorizing elements that share only a distant resemblance (e.g., the autistic spectrum), before discovering what possibly unifies them mechanistically, anticipates, distorts, or even fabricates the contours of the object being described. Both the transdiagnostic approach (e.g., integrating of autism into neurodevelopmental disorders (Michelini et al., 2024) and the dimensional approach (e.g., viewing autism as an extreme of a dimensional variation of socialization) face challenges in selecting and labeling the dimensions they favor. These choices are actually more arbitrary than the boundaries of a natural category, which are based on the averaged characteristics of a group of individuals identified by experts who have encountered a large number of them. At best, the dimensional approach accurately identifies the center of a prototype: the surface of an island may vary with ocean levels, but not the coordinates of its highest point. At worst, it includes elements that are arbitrary or irrelevant to the category. Dimensions choices are not falsifiable; they have axiomatic value in subsequent constructions, and thus have to be accepted without proof, on the basis of faith, zeitgeists, or a narrow convention on what science should be.

Diagnostic Instruments Create the Dimensionality They Pretend to Demonstrate

Each component of the dimensional autism reduction (e.g., a behaviors Autism Diagnostic Observation Schedule sign [ADOS]) contributes equally to ADOS diagnostic threshold. However, assigning equal weight to individual criteria or specific means that the summary score of a standardized diagnostic tool does not fully capture the nature of this complex condition, in its most typical and recognizable form. In this instrument, two scores of “1” are worth a score of “2,” when added together and used in the algorithm. The same quantitative threshold can be reached through the accumulation of a large number of mild signs, as well as by a smaller number of highly specific signs. Yet in our clinical experience, “1” versus “2” (for a different set of signs) are what differentiates people with high clinical certainty from those better explained by other diagnoses, or no diagnosis at all. If a canonical biological form is recognizable because its components are qualitatively recognizable and often grouped, its existence in an incomplete form cannot be equated heuristically with the lesser value of a variable. In the living world, a small elephant or an elephant missing a leg is not less of an elephant than a large, healthy elephant, and a rhinoceros is not a horned fraction of elephant-ness. Autistic-ness is not a continuous variable. A dynamic system cannot be reduced to the linear variables that control it (Scheffer et al., 2024). Although temperature is a continuous variable, its effects on the phases of matter are not.

A specific mention has to be made on the dimension of “profoundness” recommended by the Lancet commission (Lord et al., 2022). The degree of clinical severity, which hierarchizes services based on the needs and level of adaptive difficulties, should not be used to justify the abolition of boundaries between conditions. This is the case for most human variants that we group as “asymmetric developmental bifurcations” (ADB) (Mottron et al., 2025), together with some other stable variants of human development such as breech birth, being left-handed, or having twins. Breech delivery, for instance, is an example of a stable minority variation that we can consider as a heuristic model of the human variant known as autism (Mottron & Gagnon, 2023). Cases are categorized as frank breech (legs are extended, feet close to the baby's face), complete breech (both knees and hips are flexed) or incomplete breech (one or both feet or knees are presenting first). Each of its forms has the same natural category status as the category “breech birth” that unites them. There is no such thing as a spectrum of breech delivery, or as a dimension of breechdeliveriness.

We posit the temporal and cross-sectional nature of the boundaries between autism and what it is not are closer to that of breech delivery that to the “extreme far end of a dimension” (Koi, 2021). Subcategories of autism (e.g., Asperger-type, fragmented presentations) can themselves serve as prototypes, delineated by specific patterns of signs occurring in specific contexts or periods of the development. That a breech birth may result in the death of the infant or proceed smoothly does not change the specificity and delineation of this presentation, but justifies it to be considered with care (Mottron et al., 2023). Emphasizing the severity of autism is as detrimental to the advancement of autism science as denying it.

The Effect of Dimensional Approaches on Our Ability to Make Discoveries

The dimensional framework dismisses the complexity of behaviors and cognitions by condensing their variability into largely subjective, higher-order, classifying, concepts such as the “social reciprocity” or the “repetitive” (and, recently, the “profoundness”) dimension. As identifiers, these constructs are overinclusive and introduce artifactual heterogeneity in case ascertainment, with a paralyzing effect on modeling. The supposed heterogeneity of autism arises from:

the elevation of autism in the taxonomic hierarchy of broad atypical development: autism becomes a meta-category that covers the greater part of deviations from a normative social standard (some pathologic, but not necessarily) while keeping the name;

the excessive abstraction of diagnostic criteria, rather than focusing on qualitatively recognizable signs as would be, for example, prolonged objects fixation (Ozonoff et al., 2008). For example, the term “repetitive,” because of its level of abstraction, hinders our capacity to distinguish autistic repetition from that linked to cognitive impoverishment in intellectual disability, obsessive-compulsive disorder, the motor manifestations of Gilles de la Tourette syndrome, or even questions of anxious reassurance.

the relegation of language to a mere specifier, whose level is therefore uninformative for diagnosis. This implies that language levels and trajectory are equally informative across individuals and ages, whereas language trajectory plausibly contributes to define autistic prototype (Gagnon et al., 2021);

the inclusion of both syndromic and nonsyndromic forms within autism (Ziats et al., 2021);

the failure to account for developmental transformations through the use of singular criteria all along the lifespan.

We have painted elsewhere (Rabot et al., 2023) a pessimistic picture of scientific advances in autism over the past 20 years. The major advances have mainly consisted of eliminating causal factors, refuting therapeutic claims, and demonstrating the nonspecificity of the biomarkers put forward. Research into the neurobiology of autism has focused primarily on the syndromic fraction. It is becoming clear to geneticists and neurobiologists (see, e.g., Bourque et al., 2025; Douard et al., 2021) that research based on animal models, mostly derived from the implementation in an animal of a deleterious mutation found sporadically in the syndromic fraction, are minimally informative about the bulk of the autistic population. This research produces knowledge, but its scope of application remains limited to the carriers of the mutation that causes it. If we want knowledge that can be generalized at least to the center of the autistic category, we need to work on it.

We now have access to rich data showing that the characteristics of populations diagnosed very late are so different from those diagnosed early that integrating them into the same broad, spectral category is hazardous (Rodgaard et al., 2021a, 2021b, 2022). This clinically obvious conclusion has been recently considered by the genetic community (Zhang et al., 2025), after 2 decades of processing early and late diagnoses as a unique condition. It has been demonstrated how the difference between autism and typical comparison populations, indexed by the effect size of candidate biomarkers, has been reduced by up to 80% over the last 30 years (Rodgaard et al., 2019). These results, either indicate that these biomarkers were false, or demonstrate how much the autism category has merged into the general population. This would fulfill Constantino's (2011) aphorism that “autistic traits are continuously present in the general population,” with deleterious consequences for our ability to produce knowledge about autism, which is regrettable.

What applies on anything is informative on nothing. Affirming that everything is water, as Thales of Miletus suggested (Aristotle & Lawson-Tancred, 1998), is but a variation on the same kind of dimensional thinking, taking a shared component of a whole and elevating it to a self-sufficient cause or identification criteria. The logical fallacy has the effect of making the boundaries of the clinical entity unfalsifiable. What observations could effectively refute the belonging to a spectrum that cover every possible level of a variable?

Despite the Dimensional Drift of the Criteria, the Categorical Aspect of Autism Remains Supported

We need to take stock of the work that has been done in favor of a categorical conception of autism, defined as much by its signs as by its developmental trajectory. Before the current explosion of self- and hetero-diagnosis in adulthood, the diagnosis of autism was considered to be mostly stable over time (Guthrie et al., 2013) and to allow interrater consistency relative to other neuropsychiatric diagnoses (van Daalen et al., 2009). In genetics, the nature of their heritability distinguishes syndromic and nonsyndromic autism (Sztainberg & Zoghbi, 2016). Imaging studies are indicative of a specific pattern of inversion of hierarchical gradients of information processing (Bernhardt et al., 2025) rather than of gross structural or functional abnormalities. Taxometric and latent class analysis applied to combined cohorts and behavioral measures concluded to a strong support for a categorical structure (Frazier et al., 2023). Again, this arguably more complex, but also more clinically and scientifically relevant vision of autism is countered by an uncritical acceptance of its heterogeneity and its mechanistic, cognitive and genetic “dimensional” continuity with typical development as well as with other conditions.

Autism as a Quasi-Categorical Prototype

Each Living Object Has Its Own Kind of Boundary

The choice between different positions on the dimensional or categorical nature of autism may depend on the class of biological objects to which it is assimilated, from which we will borrow heuristic tools. If autism is considered a neurodevelopmental condition—a view dominant in the scientific community—then the same limits accepted for other developmental conditions can be applied to autism. Taking hyperactivity as a model, for example, shows that we can be more or less hyperactive, making the possibility of a discontinuous threshold questionable. Conversely, if we equate autism with asymmetric developmental bifurcations, the nature of the boundary between autism and what it is not becomes different. There is more of a collection of categories than a category of categories.

Two of us have recently shown (Mottron et al., 2025) the value of linking autism to Asymmetric Developmental Bifurcations, minority human variants that are nevertheless part of the species’ heritage: being left-handed, having twins, being born in breech. In all these situations, the choices available at a particular stage of development are polarized, and ADB-specific. The factual space separating the two poles of the bifurcation is strongly constrained by the nature of the bifurcation. For example, there are only two possibilities for the position of a fetus once it is too large to turn around: having its head up or down. The environmental context of pregnancy imposes a polarization of possible outcomes. For lateralization, the nature of the frontier between left-handed and right-handed is more diffracted, but polarization remains intact, specific to the species, and radically asymmetrical and stable. The interest of this type of comparison is to subvert the opposition between dimensional and categorical in favor of a type of frontier specific to each asymmetrical bifurcation. Even though there are several levels of lateralization, human lateralization remains quasi-categorical in nature, with the majority of humans being homogeneously right-lateralized, one-tenth of them being left-lateralized, and a minimal number showing no lateralization.

In these polarized situations, the boundaries of each developmental choice, whether majority or minority, are constrained not only by the binary nature of the possible choices, but also by their specific context. ADBs are the quintessential situation in which the dimensional approach is ineffective. We posit that a polarization of the same type explains the occurrence of autism as much as it constrains its quasi-categorical nature. According to the asymmetric bifurcation model, autism arises from the potential, at a specific point in the developmental program, for a minority choice in a polarized situation, whereby information is processed according to a social bias or as nonsocially prioritized information. Therefore, autism would represent the developmental possibility of an absence of social bias, that is, an absence of prioritizing the attributes of one's peers over information in general. This polarized choice is the most general that a child's information processing system has to deal with during its early years, in terms of orientation toward, detection of, and mastery of incoming information. This reversal of the processing hierarchy is currently the most economical way to account for both the semiology of autism (see companion paper) and the atypicalities observed in brain imaging (Bernhardt et al., 2025) and language learning (Kissine et al., 2023).

Prototype Theory and Diagnosis

According to Rosch's Theory (1978), prototypes arise from the interaction between a family of entities that share phenomenal or functional similarities (existing independently of the observer) and the observer's ability to subjectively group these entities into families. There are no real representatives of a family that possess all the characteristics that define it—such as the prototypical bird—and even if these characteristics exhibit a certain degree of variability—for example, the feathers of a species compared to the prototypical feather. Nevertheless, a group of humans exposed to a population comprising all kinds of birds with all kinds of feathers, as well as all kinds of nonavian animals, will have a very similar representation of the prototypical bird. They will be able to recognize a bird, even if they have never seen that species before.

Applying this analogy to autism (Mottron, 2021b) suggests that, even if no existing autistic child exhibits all the signs typically associated with autism, the prototype of autism among different groups of professionals will be very similar once they are exposed to a neighboring population. The prototype depends on individual exposure, even if the building of a prototype requires the identification of objective similarities. We only detect similarities between configurations to which we have been exposed. Therefore, there may be a secular drift in the prototype from the moment clinicians are exposed to a population that shifts gradually over the years. This is precisely what is happening in autism: the clinic's population of individuals with language delay and early diagnosis is gradually being replaced by a population with late diagnosis and self-diagnosis. The same thing happens in research when cohorts are influenced by this secular change in their composition. It would be misguided to see this as a sign of progress toward the truth without further reflection.

The prototype theory was developed to remedy the type of logical-mathematical categorization based on the presence of necessary and sufficient characteristics, which does not apply well to families of objects in the real world. Although the DSM has distanced itself somewhat from this type of categorization and made certain accommodations, it still retains the notion of an absolute threshold, whereas clinical reality points to a heterogeneity of levels of similarity within the current spectrum, with one part of the autistic population being extremely homogeneous for a certain period. Another part of the “spectrum” population is partially but qualitatively similar to the first, but its relationship to the former becomes less and less demonstrable as it moves further away from it.

An Overlap of Features Among Human Variants Does Not Alter the Distinct Character of Prototypical Autism

Sharing, among human variants, patterns of brain structure and genetic factors is trivial and does not alter the distinct character of autism in its prototypical form. The imprecision, lack of reproducibility, and lack of specificity of genetic and imaging results for autism (Muller & Fishman, 2018) may be intrinsically linked to the improper delineation of the studied population. Search for biomarkers, if any, may only progress at the cost of a radical change in assumptions, strategy and methodology, and in abandoning ill-founded dogma that heterogeneous groups automatically result in more generalizable findings that smaller homogeneous ones.

The fact that autism may lead to presentations of hyperactivity or anxiety does not imply a mechanistic connection or dimensional continuity between the two processes. Comorbidity does not negate the etiological and mechanistic autonomy of each condition. Brachial paralysis is a common consequence of some forms of breech delivery, but it does not help explain the cause of breech presentation, nor does it affect the stability of its prevalence as a possibility in human pregnancy.

The presence in autism of neurological and genetic features that are present in other conditions may also be due to variables that are trivially present in all humans. If autism represents a distinct developmental organization, utilizing the same elements as other typical humans, there is a risk of confusing the commonality of factors involved among different human variants with an etiological or mechanistic community among these differences.

However, even in future studies on homogenous prototypical cohorts, the autism scientific community should be prepared to discover that the type of difference distinguishing autistic individuals from the majority of humanity that they have been searching for decades may not exist. As a breech-presenting baby does not differ from one with a cephalic delivery beyond the presentation itself, an autistic child may not be “more” or “otherwise” different from a typical child beyond the autistic phenotype and brain reorganization itself. The outcome of these two ADBs may, however, range from dramatic to outstanding, and sometime is unremarkable.

Pending a New Foundation, What Rules Should Research and Clinical Practice Follow?

Research on autism often relies on cohorts of individuals phenotyped with instruments that allow the inclusion of participants far beyond what can be considered prototypical (Molloy et al., 2011). Some approaches are even more permissive, accepting late diagnoses or self-diagnosed individuals as representative cases.

The drift resulting from the use of such heterogeneous populations has led the field to internalize epistemological biases. Through the cumulative effect of meta-analyses, for instance, information from diverse and often incomparable sources becomes condensed and reified, transforming heterogeneity itself into an ontological “truth” for the scientific community, clinicians, and the general public. Such fallacies can introduce significant biases in addressing critical questions, such as sex-related differences in the presentation of autism. This is particularly the case when meta-analyses include studies based on self-diagnosed individuals and/or broad diagnostic criteria (Edwards et al., 2024). This methodological choice risks attributing to prototypical autism the properties of individuals who are in fact radically distinct from it, and conversely applying to these heterogeneous groups features that are specific to prototypical autism.

We therefore question the validity, for prototypical autism, and the generalizability of major conclusions drawn from individuals on the broad autism spectrum or who self-diagnose as autistic. Nevertheless, a new generation of clinically grounded scientific studies will require the shared acknowledgment of a prototypical autism category, or at least a prototypicality index, to enable more meaningful comparisons, whether approached dimensionally or categorically.

The first assumption in this direction implies the existence of an internationally and clinically consensual set of signs known to characterize prototypical autism. A cohort integrating these signs, together with a high level of clinical certainty, would provide a primary framework for identifying temporal, intrinsic, or contextual features that distinguish prototypical autism from potential phenocopies. For currently available public cohorts, mapping prototypical signs onto their best correspondence within existing standardized diagnostic tools, and extracting a subsample that meets a sufficient threshold of prototypical features, represents a reasonable approach. In the meantime, pending the formal identification of prototypical signs by expert consensus, it remains possible to estimate the prototypicality of autistic individuals based on the degree of clinical certainty associated with their diagnosis, which constitutes a provisional and nonformalized indicator of prototypicality.

In the current absence of such consensus, some authors have suggested useful signs or strategies to address prototypicality. For example, Rodgaard et al. (2024) proposed a subset of ADOS items associated with higher clinical certainty, while Gagnon et al. (2022) identified ADI-R signs that tend to co-occur in regressive autism. These lists partly overlap with the provisional set of prototypical signs proposed by Mottron and Gagnon (2023). Such a strategy, based on an estimated level of prototypicality, was applied to the ABIDE brain imaging cohort. The reanalysis of imaging data using an index of prototypicality (the verbal/nonverbal intelligence ratio) increased the signal of group differences (Hong et al., 2023; Rodgaard et al., 2024).

With regard to the reworking of the prototypical signs themselves, we propose using the expertise of clinicians exposed to an enriched population to redefine the criteria for prototypical autism. We have recently demonstrated the effectiveness of this approach through the LLM analysis of over 4,000 descriptions (justification for the reference for an autism assessment, or lay-language diagnostic reports) of more than 1,000 children (approximately half of whom received an autism diagnosis from experts). The DSM 5 signs in area B, and more specifically three of them (B1, B2 and B4), were strongly associated with a positive diagnosis by a multidisciplinary team, over sociocommunicative signs. The sentences that predicted the final expert diagnosis, and that were identified prior to the use of standardized instruments or the application of a DSM grid, described qualitatively recognizable and unitary behaviors centered on perception (Stanley et al., 2025).

We therefore recommend suspending the diagnostic application of the DSM-5 clinical specifiers. For language, the developmental language trajectory should be considered as part of the differential diagnosis. For comorbidity, the degree of clinical certainty should be weighed in the presence of a “syndromic” presentation. For cognitive developmental levels, verbal/nonverbal intelligence ratios should be incorporated into the diagnostic reasoning. In the case of adults, we recommend adding an external informant to the anamnesis in instances of self-diagnosis, and requiring signs of atypicality that are qualitatively consistent with clinical judgment during development to support the diagnosis.

This approach is preferable to relying on overarching, uncharacterized, underconceptualized and dimensional “spectrum” cohorts. The autism spectrum currently extends indefinitely toward “incomplete” or attenuated forms, such as the broader autism phenotype, and even individuals who lack most of the defining features of prototypical autism. This is precisely why the time has come to isolate within this spectrum a natural category that matches clinical certainty. Only then will we be able to determine whether nonprototypical forms result from the incomplete development or the transformation of prototypical forms, or alternatively from a lowered inclusion threshold in an uninformative, abstract category.

We are confident that a quasi-categorical prototype of autism will prove more fruitful for scientific inquiry and more beneficial for autistic individuals than the current “spectrum” framework, which combines an abrupt diagnostic threshold, abstractly defined signs, with the indefinite degrees of freedom authorized by clinical specifiers. Diagnosing autism as a natural category has enabled it to exist as a clinical, societal and, before the current shift, scientific entity. At best, the dimensional approach has led to the rediscovery of facts that were already known (Fusar-Poli et al., 2019), and at worst, it has silenced Kanner's and others’ initial categorical detection and led to the loss of the autism signal.