Individuals with Autism Spectrum Disorder (ASD) are thought to under-rely on prior knowledge in perceptual decision-making. This study examined whether this applies to decisions of attention allocation, of relevance for ‘predictive-coding’ accounts of ASD. In a visual search task, a salient but task-irrelevant distractor appeared with higher probability in one display half. Individuals with ASD learned to avoid ‘attentional capture’ by distractors in the probable region as effectively as control participants—indicating typical priors for deploying attention. However, capture by a ‘surprising’ distractor at an unlikely location led to greatly slowed identification of a subsequent target at that location—indicating that individuals with ASD attempt to control surprise (unexpected attentional capture) by over-regulating parameters in post-selective decision-making.

Individuals with autism spectrum disorder (ASD) have been widely reported to show atypicalities in predictive coding, though there remains a controversy regarding what causes such atypical processing. Suggestions range from overestimation of volatility to rigidity in the reaction to environmental changes. Here, we tested two accounts directly using duration reproduction of volatile and non-volatile interval sequences. Critically, both sequences had the same set of intervals but differed in their stimulus presentation orders. Comparing individuals with ASD vs. their matched controls, we found both groups to respond to the volatility in a similar manner, albeit with a generally reduced prior in the ASD group. Interestingly, though, relative to the control group, the ASD group exhibited a markedly reduced trust in the prior in the volatile trial session when this was performed after the non-volatile session, while both groups performed comparably in the reverse session order. Our findings suggest that it is not the learning of environmental volatility that is compromised in ASD. Rather, it is their response to a change of the volatility regimen from stable to volatile, which causes a highly inflexible weighting of prediction errors.