Autism, Neanderthal Ancestry and Exclusion

Research question and concept: Nathan Organ.

Evidence review and hypothesis paper | Version 0.2 | 4 November 2026. This is an AI-assisted working paper prepared for author review, not a report of completed original experiments or external peer review.

Abstract

Nathan Organ proposes that some cognitive characteristics now associated with autism may have been prominent in Neanderthals, whose displacement by better-connected Homo sapiens could represent an early form of coalition-driven exclusion. This paper evaluates that proposal alongside a contemporary question: can social selection against atypical contributors reduce a community’s usable cognitive capacity? A targeted narrative review examines comparative neuronal evolution, autism genetics, archaic ancestry, Neanderthal population structure, social evaluation and collective judgement. The literature supports several components independently: autism involves heterogeneous genetic pathways; some archaic variants have reported associations with autism; network asymmetry can generate competitive displacement in models; and social presentation can affect willingness to engage with autistic people. The missing links are a reliable Neanderthal cognitive phenotype, evidence of deliberate exclusion, and evidence of a resulting inherited cognitive decline. Recent evidence also challenges a uniform picture of isolated Neanderthal communities. The paper therefore separates three outcomes: inherited cognitive potential, individual performance under particular conditions, and the knowledge a group successfully accesses and uses. It develops a coalition-filtering hypothesis in which useful contributions can be lost through reputation, communication mismatch or exclusion, without any genetic deterioration. Proposed experiments test content evaluation, contribution uptake and decision accuracy under different communication and governance arrangements. The ancient hypothesis remains open but weakly constrained; the contemporary mechanism offers a more directly testable NableU research programme.

Keywords: autism; Neanderthal ancestry; cognitive diversity; social exclusion; collective intelligence; evolutionary trade-offs; cognitive equity.

1. The research question

The initiating hypothesis is: perhaps Neanderthals possessed cognitive tendencies resembling some autistic profiles, while competing Homo sapiens populations had greater capacity to organise alliances, control access to resources and marginalise outsiders. Organ describes the proposed process as the “original cancel culture”. The further concern is that success at coalition-building may have displaced valuable independent cognition and left subsequent societies less capable than they could otherwise have been.

A useful research paper must preserve this proposed mechanism while separating its claims. Evidence that a genetic variant contributes to autism does not identify the behaviour of the ancient population carrying it. Evidence that a population was small does not identify why it was small. Evidence that a group excludes outsiders does not establish that the outsiders were more accurate. Each transition requires its own test.

For this paper, coalition-driven exclusion means that access to participation, resources or influence depends partly on social alliances and reputation rather than only on the relevant contribution. “Cancel culture” is retained as Organ’s originating metaphor, not used as an archaeological diagnosis. Competitive displacement, disagreement, punishment of exploitation and organised reputational exclusion are distinct processes.

The strongest version of the proposal is conditional: when social selection rejects genuinely useful cognitive contributions for reasons insufficiently connected to their quality, a community may lose functional capability. Its ancient extension proposes that something analogous contributed to Neanderthal displacement. Its strongest biological extension proposes a lasting reduction in inherited cognitive potential. The latter two propositions require substantially more evidence than the contemporary mechanism.

Modern autistic people are Homo sapiens. Archaic ancestry, present-day diagnostic categories and particular cognitive abilities are different analytical levels; treating them as interchangeable would make the theory impossible to test reliably.

2. Method and evidential standards

This is a targeted critical narrative review, with searches and source checking completed on 4 November 2026. Searches combined autism, Neanderthal introgression, neuronal evolution, population structure, network competition, social exclusion, information transfer, collective intelligence, brain size and cognitive-score trends. Primary articles and their publisher, PubMed, PubMed Central or author-repository records were prioritised. Studies contradicting the initial hypothesis, larger replications and recent findings were deliberately included.

The review includes 28 research publications, including empirical studies, computational models and a published reanalysis debate. It is not a systematic review: there was no exhaustive database export, registered review protocol, duplicate screening, formal risk-of-bias instrument or meta-analysis. No participant data, ancient genomes or supplementary datasets were independently reanalysed. Some sources were available only through their abstracts or indexed primary-source text; access depth is recorded in the accompanying evidence register. Citation counts and journal prestige were not treated as proof.

Four descriptions are used throughout: observed for a study’s bounded result; inferred for an interpretation requiring assumptions; modelled for consequences of specified conditions; and proposed for a mechanism awaiting direct testing. A study can provide good evidence for a narrow result while providing little evidence for the larger evolutionary account.

3. Autism and the evolution of human cognition

3.1 A plausible developmental trade-off

Starr and Fraser’s 2025 analysis compared single-nucleus gene-expression datasets encompassing more than one million neurons across six mammals. Human layer 2/3 intratelencephalic excitatory neurons showed unusually rapid expression evolution, including reduced expression of autism-associated genes. The authors interpreted the pattern as consistent with selection for changes that benefited ancestors while increasing susceptibility to developmental perturbation. This supports an evolutionary trade-off hypothesis. It neither identifies the particular cognitive benefit nor compares living Neanderthal neurons with sapiens neurons. A human-versus-other-primate contrast also cannot, by itself, date a change relative to the Neanderthal–sapiens split, or establish whether the relevant changes were shared by both lineages. [1]

Consequently, the video’s suggestion that autism “made us smart” compresses several untested transitions. A gene can contribute to normal development and to a clinical outcome under different conditions. Selection on its regulation does not demonstrate selection for a modern diagnosis. The relevant research question concerns shared developmental machinery and its trade-offs, rather than whether an entire diagnostic category was selected as a single ancestral adaptation.

3.2 There is no single genetic origin to reconstruct

Grove and colleagues’ large autism genome-wide association study identified common-variant associations and genetic overlap with educational attainment and intelligence. Such correlations concern population-level statistical architecture; they do not establish an intelligence advantage for every autistic person, or show that the associated variants improved ancestral survival. [2]

Other findings require a broader account. Warrier and colleagues found differing relationships between autism-related polygenic scores, rare variants and co-occurring developmental disabilities. Zhou and colleagues’ analysis of 42,607 autism cases identified both inherited and de novo contributions, including moderate-risk genes. Litman and colleagues subsequently derived four phenotypic classes from 5,392 participants and identified differing genetic patterns. Those classes are model-based results, not a final division of autism into four natural kinds. Together, these studies support substantial heterogeneity rather than one uniform cognitive strategy. [3–5]

An evolutionary explanation must therefore accommodate potentially advantageous traits, neutral variation, developmental vulnerabilities, newly arising variants and substantial disability. A trait’s persistence does not, on its own, prove that it benefits the group. Nor does a benefit in one task imply an overall fitness advantage. The hypothesis should be evaluated at the level of specified traits and mechanisms, not through a universal “autistic brain” profile.

4. What Neanderthal evidence establishes

4.1 A genetic connection worth investigating

Pauly and colleagues reported enrichment of a subset of Neanderthal-associated variants in autistic participants, using SPARK, GTEx and 1000 Genomes data. They did not find a general excess of Neanderthal DNA in autistic people. Some signals also occurred in unaffected siblings. The study provides a candidate connection, not a diagnostic reconstruction of Neanderthals. The authors acknowledge population-structure and control-matching limitations; shared ancient variants can also complicate attribution to introgression. Independent, ancestry-aware replication is therefore a priority. [6]

There is a further inference problem even if the associations replicate. A variant’s association in a contemporary genome does not determine its effect in a different genetic, developmental and ecological background. Wei and colleagues’ broader analysis illustrates why matching allele frequency and linkage structure, and separating introgressed signals from linked modern variants, are important in archaic-trait research. [7]

The defensible claim is that archaic inheritance may contribute to some present-day variation relevant to autism. The reviewed evidence does not establish that Neanderthals were generally autistic, that autism originated exclusively with them, or that the associated variants carried an ancestral cognitive advantage.

Skov and colleagues analysed 13 Neanderthals from two Altai caves, identifying close kinship and evidence consistent with small communities and female movement between them. Slimak and colleagues’ “Thorin” genome from France indicated a lineage with approximately 50,000 years of genetic isolation. These findings make restricted connectivity plausible in particular populations. They do not diagnose autism or explain whether isolation reflected geography, demography, social preferences or exclusion. [8,9]

Importantly, Bossoms Mesa and colleagues’ June 2026 study examined 27 late Neanderthals from ten sites in northwestern Europe. Their findings suggested larger or better-connected populations than some Altai comparisons, without evidence of an increasing genetic load through the sampled period. A species-wide account of socially disconnected, progressively deteriorating Neanderthals is therefore too simple. Regional and temporal variation must be built into any competition hypothesis. [10]

Archaeology adds evidence of organisation and technical competence. The Bruniquel Cave constructions, dated to about 176,500 years ago, indicate purposeful activity deep underground. Analysis of ochre–bitumen adhesives at Le Moustier supports sophisticated material processing. Neither result supplies comparative IQ scores or identifies a neurotype, but both constrain any account of Neanderthal cognitive and social capabilities. [11,12]

4.3 Network advantage and individual intelligence are different questions

Keskin, Özgür and Sağlam modelled how differences in starting social-network structures could shape competition between populations. In their model, a larger, better-connected sapiens network can produce Neanderthal extinction. This demonstrates that network asymmetry can be sufficient under specified assumptions. It does not establish that the assumed asymmetry always existed, or that reputational exclusion, autism-like cognition or greater sapiens intelligence caused it. [13]

Alternative models also achieve extinction without cognitive inferiority. Vaesen and colleagues modelled small-population vulnerabilities, including inbreeding, low-density effects and demographic stochasticity. Their results show that several mechanisms can generate a similar endpoint. Models of sufficiency are not unique historical explanations. [14]

Finally, genomic evidence documents interbreeding and surviving Neanderthal ancestry. Disappearance as a distinct population was not complete genetic erasure. Any proposed loss of particular capacities must explain which variants or developmental combinations were lost, which persisted, and how their effects could be established. [15]

The ancient hypothesis can be investigated, but its decisive evidence is missing: a reliable cognitive reconstruction, a demonstrated exclusion mechanism and a link between exclusion and the loss of specified capabilities.

5. How strengths can become liabilities

The phrase “strengths that became weaknesses” can describe several different processes. A characteristic may be beneficial in one task and costly in another. An environment may impose demands that exceed available support. A useful contribution may be socially misjudged. These processes should not be collapsed into a claim that every impairment conceals a special talent.

There are bounded examples of cognitive advantages. O’Riordan and colleagues found better performance by autistic children on difficult visual-search tasks in two experiments. Farmer, Baron-Cohen and Skylark found more consistent choices in autistic adults in a task involving context-dependent preference reversals. These results support investigating specific forms of discrimination and decision consistency. They do not establish universal perceptual superiority, truthfulness, rationality or resistance to social influence. [16,17]

The following pathways are proposed task–environment interactions, not established descriptions of all autistic people or of Neanderthals.

Fine-grained discrimination. A person may detect an anomaly that others overlook. The same attention becomes costly when irrelevant detail competes with task priorities or the environment demands rapid filtering.

Stable evaluation criteria. Consistency can resist irrelevant framing. It becomes a liability when relevant conditions change and updating is required, or when negotiation treats any resistance as unacceptable.

Sustained specialisation. Concentrated effort can build detailed knowledge. Frequent switching, interruptions or exclusion from the relevant work can prevent that knowledge from being used.

Explicit, direct communication. Direct statements can expose assumptions and errors. Their value may be lost when listeners judge delivery as a social violation before examining the claim.

Independent judgement. A genuinely informative minority estimate can improve a decision. Exclusion can remove its route into the group, although independence can also preserve error. The test is accuracy and relevance, not dissent by itself.

The important unit of analysis is the complete interaction: person, task, environment, support and social response. A signal that is valuable for identifying a fault can be costly in a setting organised around rapid agreement. That proposition concerns institutional design; it does not require the contributor to be correct on every occasion.

This creates two separate NableU questions. What support allows a person to exercise an ability without unsustainable cost? What evaluation process allows others to recognise a useful contribution without requiring its author to imitate a preferred social style? Both can be investigated directly, without settling the evolutionary story.

6. The proposed coalition-filtering mechanism

6.1 A measurable contemporary pathway

Sasson and colleagues found less favourable first impressions and reduced willingness to interact with autistic people across three studies. In their modality comparison, the disadvantage was not found when observers assessed speech transcripts without audiovisual cues. This provides evidence that presentation can influence social reception independently of the conversational content available in that condition. It does not establish that rejected contributions were objectively better, or that every real-world exclusion follows this pathway. [18]

The proposed extension is a causal chain: an atypical but informative contribution attracts a negative social judgement; that judgement reduces access or influence; the group fails to use relevant information; decision quality deteriorates. Each link must be measured. Evidence for the first link does not automatically establish the last.

Communication differences also resist a simple deficit story. Crompton and colleagues’ 2020 study found effective autistic-to-autistic information transfer and a mixed-neurotype disadvantage. A larger 2025 registered report with 311 participants found no information-transfer difference between autistic, non-autistic and mixed chains, although rapport differed. The later result challenges both a general autistic communication deficit in that task and an inevitable mixed-neurotype penalty. Comfortable interaction and accurate information transfer are distinguishable outcomes. [19,20]

A viable study must therefore ask whether accurate information was transmitted, whether it was understood, whether it influenced the decision, and how costly the interaction was for participants. Rapport alone cannot answer those questions.

6.2 Coordination has benefits as well as costs

Lorenz and colleagues experimentally showed that social influence can reduce diversity of estimates without a corresponding improvement in accuracy. Becker and colleagues demonstrated that interaction can instead improve collective estimates under suitable network arrangements. The implication is conditional: the quality of influence and network structure can determine whether coordination helps or harms judgement. [21,22]

Exclusion can also serve legitimate cooperative functions. Feinberg, Willer and Schultz found that gossip and ostracism could promote cooperation in experimental groups. A research account should therefore distinguish sanctions against exploitation from sanctions against unfamiliar but useful contributions. Treating every form of norm enforcement as a cognitive loss would ignore a central counterargument. [23]

Organ’s proposal is strongest where the social filter and the task’s truth conditions come apart. A group may reward confidence, affiliation or familiar delivery while underweighting verifiable evidence. However, dissent is not evidence of correctness, and popularity is not evidence of error. The relevant comparison is contribution quality and group outcome under alternative decision procedures.

6.3 An illustrative mathematical consequence

Suppose ten people independently estimate a quantity. Assume equally reliable, unbiased estimates, each with error variance σ². If all pairs have error correlation ρ, the variance of their equally weighted mean is:

Var(mean error) = σ² × [1 + (n − 1)ρ] / n.

For n = 10, independent errors give 0.10σ². If pairwise error correlation is 0.80, the variance becomes 0.82σ². The individual error variance has not changed, but the benefit of aggregation has largely disappeared. This follows directly by expanding the variance of a sum, with n variance terms and n(n − 1) covariance terms.

This is an illustrative calculation, not a measured autism effect or a reconstruction of prehistory. Real contributors differ in expertise, bias and reliability; informative social learning can improve them. The calculation establishes only that collective performance can deteriorate through correlated errors while individual capacity remains unchanged. It motivates measuring both contribution quality and independence rather than counting agreement as success.

7. What “lower cognitive capacity” could mean

7.1 Three outcomes must be separated

Inherited potential concerns biological capacities and their developmental possibilities. A claim that ancient exclusion reduced it requires evidence of relevant heritable variation, differential reproduction or disappearance, and a defensible connection to the target ability. The reviewed studies do not provide this chain.

Realised individual performance concerns what a person achieves under particular educational, sensory, health, motivational and task conditions. It can change without a corresponding change in inherited potential. Research must specify the conditions and outcome rather than treating any test as an exhaustive measure of a person.

Usable collective capacity concerns the knowledge and abilities that a group can actually mobilise. The proposed coalition filter can reduce it by excluding a knowledgeable contributor, ignoring an accurate warning, or making contribution prohibitively costly. This is the most directly testable interpretation of Organ’s concern. It remains an empirical question whether a given exclusion reduces net decision quality after coordination costs are considered.

7.2 Recent score declines do not identify an ancient genetic cause

Bratsberg and Rogeberg’s 2018 analysis of Norwegian male conscripts found that the rise and subsequent fall in cognitive scores could be recovered within families, supporting environmental explanations over changing family composition as the main driver in those cohorts. The design does not settle every population or every possible small genetic effect, but it weighs against interpreting those trends as evidence of a simple inherited decline. [24]

A 2026 study of 579,379 Norwegian men found that score trends differed by paternal income and were consistent with educational expansion temporarily offsetting an underlying decline in performance. Its results connect the interpretation of recent trends to schooling and wider conditions. They do not connect those trends to prehistoric ostracism or establish a universal decline in humanity’s genetic capacity. [25]

The paper therefore does not take “people are now less intelligent” as an established premise. It identifies specific observed score trends and a separate hypothesis about wasted collective ability. Explaining one through the other would require an additional causal study.

7.3 Brain volume does not close the gap

A proposed recent decrease in human brain size has itself generated disagreement. DeSilva and colleagues argued for a Holocene reduction; Villmoare and Grabowski challenged the inference from the dataset and analysis; DeSilva and colleagues subsequently defended a reduction while discussing its timing and explanation. This debate does not provide a demonstrated chain from social exclusion to diminished intelligence. Skull-volume trends cannot substitute for the missing behavioural and genetic evidence. [26–28]

8. A testable NableU research programme

Recruit autistic and non-autistic contributors to tasks with independently assessable answers, such as identifying flaws in a service plan, comparing evidence or detecting an inconsistency. Preserve each contribution’s substance and randomly assign independent evaluators to suitable presentation conditions, such as a transcript, audio or video. Use both carefully standardised material and a separate naturalistic task, because controlling delivery can otherwise remove the very phenomenon being studied.

Measure objective correctness, perceived credibility, willingness to collaborate and actual uptake of advice. Record communication preferences and relevant task expertise. A diagnostic label should not be used as a proxy for expertise. Transcript quality and information lost between modalities require explicit checks; source identity, content and presentation must not be inadvertently confounded.

Prediction: for some contributions, presentation changes uptake after correctness is held constant. Evidence against the mechanism: adequately precise estimates show no meaningful presentation penalty, or the apparent penalty disappears when genuine task-relevant differences are controlled. This study would test contemporary reception, not an ancient cognitive phenotype.

Study B: Test whether exclusion reduces group accuracy

Compare group decision procedures using the same pool of contributions. One procedure allows ordinary discussion and informal influence; another obtains independent initial judgements, provides accessible written evidence, and evaluates disputed claims against predefined criteria. Include tasks where a minority contribution is correct and tasks where it is wrong. Otherwise, the experiment would reward dissent by construction.

Measure final accuracy, calibration, error correction, contribution uptake, decision time and participant effort. Social comfort and perceived inclusion are additional outcomes, not substitutes for task performance. Randomisation must operate at the appropriate group level, with analysis accounting for participants nested within groups and repeated tasks.

Prediction: content-focused procedures improve accuracy where an informative minority would otherwise be discounted. Evidence against the mechanism: exclusion has no measurable accuracy cost, the protected contributions are generally uninformative, or any benefit is outweighed by coordination costs under the specified conditions. An inclusion benefit without an accuracy benefit would still be a meaningful finding, but a different one.

Study C: Evaluate practical participation changes

Co-design a pilot within a voluntary research setting: asynchronous contributions, participant-controlled communication channels, clear reasons for moderation, an evidence review route and protection from retaliation for good-faith correction. Compare outcomes against a prespecified baseline or randomised rollout where feasible. The intervention should not assume that written communication suits everyone; accessible alternatives, including supported communication, are essential design requirements.

Assess whether participants can contribute, whether accurate contributions are used, and whether effort and withdrawal change. Avoid a composite success score that conceals poorer accuracy behind improved satisfaction. Separate research consent from service access and account standing. No participant should have to accept genetic testing or a speculative ancestral identity to receive support.

Study D: Investigate the ancient extension independently

A genetics collaboration could preregister replication of specific archaic-variant associations using genetically informed ancestry adjustment, harmonised controls, within-family analyses and appropriate multiple-testing correction. Fine-mapping and functional work would then be needed to identify causal effects. Even successful replication would leave the ancient behavioural phenotype unresolved.

A palaeoanthropological collaboration could test region-specific network predictions against dated movement, material-exchange and genomic evidence, rather than treating a single isolated lineage as representative. Alternative demographic, ecological and assimilation models should be fitted to the same evidence. A model earns support by predicting observations better than competitors, not merely by being capable of producing extinction.

Before any participant study, specify the smallest effect worth detecting, simulate statistical power, preregister primary outcomes and obtain appropriate independent ethics review. No sample size, approval, funding or results are claimed here. Participants with varied support needs should help shape the research; any sampling exclusions must be justified and reported. Research data should be minimised and kept separate from operational participant records.

9. Claim-by-claim assessment

Human neuronal evolution altered susceptibility to autism. This is a plausible interpretation of comparative gene-expression evidence. Functional testing and better dating of the evolutionary changes would strengthen it. [1]

Some Neanderthal-associated variants contribute to autism-related variation. This is a candidate association, not a complete account of autism’s origins. Independent ancestry-aware replication and causal fine-mapping are needed. [6,7]

Neanderthals generally had an autistic cognitive profile. This is not established. Ancestry associations, isolated communities and technical artefacts do not independently reconstruct a modern diagnostic profile.

Network structure can contribute to competitive displacement. This is a conditional modelling result. Historical support requires region-specific predictions and comparison with alternative models. [13,14]

Deliberate reputational exclusion caused Neanderthal disappearance. This remains speculative. No direct evidence distinguishing that mechanism from other competition, ecological constraints, demographic processes or isolation was identified in this review.

Contemporary social filtering can reduce usable collective capacity. This is a plausible, testable synthesis. Components have empirical support, but experiments must connect reception and exclusion to objective group outcomes. [18,21,22]

Ancient exclusion caused a lasting decline in inherited human intelligence. The reviewed evidence does not establish this. It would require a demonstrated selection pathway, loss of relevant functional variation and defensible cognitive effects.

10. Limitations and conclusion

The ancient component is constrained by sparse and uneven sampling, ambiguous behavioural proxies and the difficulty of reconstructing cognition from genomes and artefacts. The modern component is constrained by task specificity, selective recruitment and incomplete coverage of people with substantial support needs. Findings from brief social evaluations, laboratory estimates or one country’s conscription tests cannot be generalised to every institution, neurotype or historical period. These limitations identify where the theory needs better tests; they are not reasons to abandon investigation.

The paper also avoids making dignity conditional on a demonstrated advantage. NableU can investigate cognitive contributions while supporting people whose impairments remain substantial under supportive conditions. A useful theory should improve access and understanding without imposing a requirement to be exceptionally productive.

The central question concerns selection: what happens when the ability to gain social acceptance is rewarded more reliably than the ability to improve a decision? Evolutionary success, institutional influence and epistemic accuracy are distinguishable outcomes. The evidence reviewed here supports investigating their divergence, but does not show that Neanderthal disappearance was the first instance or that modern humanity is genetically less capable because of it.

The resulting hypothesis is precise: under identifiable conditions, coalition-based evaluation may suppress accurate, non-redundant contributions and thereby reduce usable collective capacity. This can occur while individual abilities remain intact. Its practical implication is to test systems that preserve cooperation while making useful differences easier to recognise, evaluate and use.

For NableU, the immediate research priority is therefore the relationship between contribution, reception and outcome. The Neanderthal hypothesis remains a distinct, longer-range line of inquiry. Keeping these lines separate protects the original question from claims the present evidence cannot carry, while turning its contemporary mechanism into something measurable.

Research integrity and provenance

The initiating hypothesis and research direction are attributed to Nathan Organ. Proposed interventions are research recommendations, not existing NableU policy. Numbered citations correspond to the Sources and further reading section. The editorial evidence register records source-access limits; subsequent corrections should retain version history.

Sources and further reading

NableU Community Systems Incorporated advocates for dignity, agency, autonomy, cognitive equity and cognitive sovereignty. About NableU · Back to the article title