The autistic adult brain keeps the lights on: how uncertainty is handled differently

A long-standing hypothesis holds that the autistic brain underweights expectation and overweights incoming sensory evidence – the world stays surprising even when it should become predictable. This study from the MEG-Center at Moscow State University of Psychology and Education puts that idea to a direct neural test in adults, a population still badly underrepresented in autism neuroscience. The design is elegant precisely because it is undramatic: a two-choice value task with stable 70:30 reward probabilities, repeated across five blocks while stimulus pairs and reward values changed. Nothing forces a failure. The question is not whether autistic adults can learn, but what their brains do once learning has happened.

Behaviourally, the two groups were hard to tell apart. Both improved, both extracted the probabilistic rule, both correctly rated the high-probability option as the better bet. The computational modelling, however, showed that autistic participants held their prior beliefs with lower precision – their internal model of the task was less confidently weighted, even when their choices were correct. This is the quiet signature the behavioural data would have hidden.

The neural picture is where the two groups diverge sharply. In neurotypical adults, learning bought efficiency. As beliefs firmed up across blocks, the brain spent less: pre-decision alpha-beta suppression in decision-relevant cortex steadily declined, and responses to feedback shrank as the feedback became redundant. This is what predictive coding expects – a confident system stops working hard to explain what it already anticipates. Autistic adults broke this coupling. Their cortex did not power down as competence grew. Trial by trial, even when belief precision was high, neural activation stayed elevated. Certainty and effort, normally inversely linked, were decoupled.

The authors read this as atypically rigid and enhanced allocation of neural resources to advantageous decisions – a brain that treats a settled, well-learned probabilistic environment as if it still demanded full vigilance. They frame it as a possible aversive response to the irreducible uncertainty baked into any probabilistic task: even at 70:30, you will be wrong three times in ten, and that residual unpredictability may never get filed away as tolerable.

The two findings fit together. Lower belief precision means the internal model is held less confidently, and a less confident model gives the cortex less licence to suppress its own activity – there is always a little more of the environment left to explain. In neurotypical adults the loop closes: confidence climbs, prediction error falls, effort drops. In autistic adults the loop stays open. Crucially, this is not a learning deficit dressed up in neural language; the behaviour was indistinguishable. It is a difference in the economics of certainty – how readily the mature brain is willing to declare a problem solved and stand down.

For clinicians, the value here is conceptual rather than diagnostic. It reframes autistic exhaustion in ordinary, "manageable" situations not as a failure of skill but as a metabolic cost of never getting to coast. The behaviour looks fine; the brain is paying full price for it.

What predictive coding actually predicts

Predictive coding casts the brain as a prediction machine that suppresses its own activity once the world matches expectation. The neurotypical pattern – less cortical effort as confidence rises – is the textbook case. The autistic pattern is its mirror image: confidence rises but the machinery never relaxes, as if every trial were still informative.

Why an adult sample matters

Most autism neuroscience is done in children, where development confounds everything. Studying high-functioning adults who have already adapted to daily life isolates a stable trait-level difference in how the mature brain manages predictability – the kind of person who, in clinic, reports being "fine but drained" by routine demands.