When Prediction Replaces Perception

Human beings do not necessarily perceive one another in the same way. Nor are people likely to be perceived by others as they perceive themselves. The implications of perception become profound when considered alongside the recent neuroscience on social cognition. If (neuro)divergent cognitive architectures use different mechanisms when processing interactions, then differences may reflect something more fundamental: a divergence in the relationship between prediction, perception and conscious evaluation. This is significant because perception determines experienced reality.

As explored previously within the Divergent Genomics framework, truth is what is. Experienced reality, however, is what a nervous system constructs from what is. In psychological theory, a schema is one mechanism through which those two things can differ. Schemas are internal models built from previous experience. They allow the brain to interpret new situations without processing everything from scratch. Social schemas extend this to people, roles and behaviour: teacher, student, doctor, criminal, authority figure, trustworthy person, disabled person. They are efficient, but they also create assumptions. Once an interpretation is expected, incoming information can be processed through that expectation. What feels like accurate perception may therefore contain substantial conjecture.

Feng et al. (2026) examined social interaction between people with high and low autistic traits using fNIRS hyperscanning, which measures cortical activity simultaneously across interacting individuals. During active discussion, low-autistic-trait pairs showed greater inter-brain synchronisation involving the right temporoparietal junction (rTPJ), while high-autistic-trait pairs showed stronger synchronisation involving the right dorsolateral prefrontal cortex (rDLPFC). The synchronisation occurred in a different region of the brain altogether. This distinction becomes particularly interesting when the functions of those regions are considered alongside other experimental findings.

The rTPJ has been repeatedly implicated in the processing of discrepancies between social expectation and observed behaviour. Park et al. (2021) experimentally showed that people's existing impressions of another person influenced subsequent evaluations, while behaviour that violated those expectations produced a social prediction error associated with rTPJ activity. Greater rTPJ recruitment was associated with greater updating when observed behaviour contradicted an existing impression.

Earlier electrophysiological research provides an additional connection to inference itself. Van Duynslaeger, Van Overwalle, and Verstraeten (2007) compared intentional and spontaneous trait inference and found greater temporoparietal recruitment during spontaneous inference following information inconsistent with an existing trait expectation. Their behavioural and electrophysiological findings indicated that people had already generated trait inferences about another person even when forming an impression was not their explicit task. This means that social information does not arrive to the person without interpretation already attached to it. Previous expectations can be established, behaviour can be compared against those expectations, and person-level meaning can be constructed spontaneously.

The rDLPFC, in contrast, is strongly involved in working memory, cognitive control and the active maintenance and processing of information. Feng et al.'s (2026) findings demonstrate that successful active interaction can recruit different neural architectures, with stronger coordination occurring in a region associated with active cognitive control among the high-autistic-trait pairs.

Chambon et al. (2017) experimentally manipulated both prior expectations and presently available movement information while participants judged other people's intentions. Autistic adults were less influenced by prior social expectations. Importantly, they remained capable of learning statistical regularities from observed behaviour, allowing previous expectations to be modified through experience. The experiment itself demonstrates something directly relevant here: present evidence and prior expectation can be experimentally separated, and their relative weighting significantly differs in autistic cognition.

Predictive-processing research points in the same general direction. A comprehensive review by Angeletos Chrysaitis and Seriès (2023) found heterogeneous results across tasks and domains. Recent experimental work instead suggests differences in the dynamics through which priors and sensory evidence are weighted and updated. Shi et al. (2025) reported greater initial reliance on sensory information and reduced initial weighting of priors in autistic participants, followed by changing prior influence across repeated updating. The relevant distinction may therefore be dynamic rather than absolute.

These findings suggest that autistic cognition may keep presently available evidence comparatively influential for longer, while other cognitive architectures may resolve ambiguity more rapidly through previously learned expectations.

The important variable may be how quickly uncertainty is converted into assumed meaning. A prediction-heavy system can resolve ambiguity rapidly, whereas an actively evaluative system keeps the question open. The latter may be cognitively expensive, but keeping a question open also keeps contradictory evidence available.

Schemas are not built only from private experience; they are also transmitted socially. Families, schools, workplaces, institutions, peer groups and cultures teach people what particular behaviours are supposed to mean, who should possess authority, which behaviour is acceptable and what deviation looks like.

There is neural evidence that disagreement with a group can itself operate through prediction and learning mechanisms. Klucharev et al. (2009) found that when an individual's judgement conflicted with group opinion, the discrepancy produced a neural response resembling a reinforcement-learning prediction-error signal. The magnitude of that response predicted subsequent movement of the individual's judgement towards the group.

Difference from social consensus itself can become a learning signal. There are several distinct layers to this:

• Reality - what is actually occurring.

• Perception - information available to the organism.

• Schema-mediated interpretation - the model used to interpret that information.

• Socially transmitted schemas - models partly supplied and reinforced by the surrounding culture.

This provides a cognitive mechanism through which social conditioning can operate. A society does not have to determine every person's conclusions individually if it can establish the assumptions through which evidence is habitually interpreted.

This also intersects with deindividuation - the reduction of individual distinctiveness within group contexts - described by Festinger, Pepitone and Newcomb (1952), and horizontal policing: the peer-to-peer monitoring and enforcement of socially established norms. This peer enforcement is reflected in the literature on injunctive norms, where violations of socially approved expectations can elicit sanctions from other group members, including disapproval, rejection and ostracism.

Structured social environment such as a school/institution

↓

Repeated expectations about authority, roles and behaviour

↓

Shared social schemas

↓

Group norms

↓

Deviation becomes immediately noticeable

↓

Peers correct the deviation

↓

External enforcement becomes horizontal

↓

The individual anticipates the social cost and begins policing themselves

With horizontal policing, the teacher no longer needs to say “Don't question it.” Another student says “Why do you always have to argue?” and eventually nobody needs to say anything. The individual thinks I probably shouldn't say anything and eventually maybe I'm the problem. The conditioning has then migrated from school/institution → group → individual consciousness.

What happens if a cognitive architecture is less automatically captured by some of those socially transmitted expectations?

It may produce exactly the kinds of responses repeatedly labelled socially unusual:

“Why? that rule doesn't make sense.”

“But that contradicts what you said yesterday.”

“Why does their position make them correct?”

“Everyone believing something doesn't make it true.”

“Why do I have to behave like everybody else?”

Those responses can appear socially disruptive precisely because the person is not automatically reproducing the expected interpretation. This is where the historical pathologization of autistic social cognition becomes uncomfortable.

Some traits commonly treated principally as problems - questioning arbitrary rules, reduced responsiveness to unsupported authority, literal interpretation, resistance to social convention, detection of inconsistency, refusal to pretend that something makes sense when it does not - could reflect reduced capture by normative social inference rather than an assumed failure of cognition or social ineptitude.

Historically, however, institutions have often treated failure to conform as evidence of failure within the individual. That creates an inversion:

The institution defines the behavioural norm

↓

Divergence from the norm is identified

↓

Divergence is individualised

↓

The individual becomes the object of correction

The institution itself disappears from the causal analysis.

The same underlying problem appears whenever variation is measured against a constructed reference condition and then described principally in terms of deviation from it, rather than asking what the variation itself represents biologically, cognitively or evolutionarily.

A healthy educational system should therefore uphold differentiation while maintaining integration: expose people to competing models, teach them to interrogate premises, distinguish evidence from authority, tolerate uncertainty, recognise their own conditioning and retain enough individuality to ask: But is that actually true?

Differences need not threaten integration. A healthy system would be capable of containing differences without suppressing them.

Autistic people have repeatedly been described as having impaired social skills. Yet Sasson et al. (2017) found that non-autistic observers formed systematically less favourable impressions of autistic people from extremely brief exposures, including reduced willingness to interact with them. That reverses the direction of the question. If an observer encounters a person, rapidly infers something negative from unfamiliar behaviour and then responds to that inference rather than the person, where exactly is the perceptual error?

A person may not receive what somebody anticipates they will. That is not the same thing as failing to perceive what is actually there. What has sometimes been labelled deficient social perception may therefore reflect reduced reliance upon, resistance to, or incompatibility with socially dominant predictive assumptions.

Majority agreement cannot settle the question; a shared assumption is still an assumption. The danger becomes greater when the same predictions are distributed across a population. One person's mistaken inference is an error. A widely shared inference can become a social reality.

People learn which facial expressions supposedly indicate hostility, which tones signify disrespect, what honesty looks like, what intelligence sounds like, what empathy should look like and how a trustworthy person should behave. Eventually those associations stop feeling like interpretations. They become general consensus, but consensus does not turn an assumption into truth.

Worse, prediction can help create the reality that later appears to prove it correct:

• Expect hostility.

• Interpret ambiguity as hostility.

• Respond defensively.

• The interaction deteriorates.

The deterioration appears to validate the original judgement, as with a self-fulfilling prophecy. Scale that process across institutions and populations and socially transmitted schemas begin shaping not only behaviour, but experienced reality itself.

That brings us back to consciousness. If different neurocognitive architectures assign different weight to prior expectation and incoming evidence, neurodivergence may provide naturally occurring variation in how readily socially supplied interpretations become incorporated into conscious reality.

What happens when a mind does not automatically reproduce the interpretation everyone else has learned to expect? A cognitive architecture that does not reproduce the majority's interpretation is not necessarily perceiving incorrectly. The person may be doing something far more disruptive: looking again.

The implications of the research herein may extend considerably further than social interaction. The mechanisms examined here - social assumptions, prior weighting, prediction error and active evaluation - are also mechanisms of learning. They operate while the nervous system itself is developing. If different neurocognitive architectures differ in how present evidence modifies what will subsequently be expected, then the implications are within neurodevelopment itself.

The areas of the brain discussed in this article mature along different developmental timelines, with the rDLPFC undergoing protracted maturation from adolescence into adulthood. If high-autistic-trait individuals achieve social coordination through greater recruitment of the rDLPFC, this raises the possibility that reliance on a later-maturing neural system could contribute to differences in the timing or trajectory of social development from infancy.

Natural biological (Neuro)divergence.

The adult findings make the developmental question testable. Traits that systems have pathologised as deficits in social skills may not be deficits at all. They may ultimately prove to be the identifiable - and otherwise beneficial - outcomes of divergent neurodevelopment.

Image: James Ensor, The Intrigue, 1890. Public domain/CC0, Royal Museum of Fine Arts Antwerp.

Contact

Reach out with questions or collaboration ideas.

Email

AChambers@divergentgenomics.org

© Alexandra Chambers 2026. All rights reserved.