Suboptimal but intact integration of Bayesian components during perceptual decision-making in autism

Background: Alterations in sensory perception, a core phenotype of autism, are attributed to imbalanced integration of sensory information and prior knowledge during perceptual statistical (Bayesian) inference. This hypothesis has gained momentum in recent years, partly because it can be implemented both at the computational level, as in Bayesian perception, and at the level of canonical neural microcircuitry, as in predictive coding. However, empirical investigations have yielded conflicting results with evidence remaining limited. Critically, previous studies did not assess the independent contributions of priors and sensory uncertainty to the inference. Method: We addressed this gap by quantitatively assessing both the independent and interdependent contributions of priors and sensory uncertainty to perceptual decision-making in autistic and non-autistic individuals (N = 126) during an orientation categorization task. Results: Contrary to common views, autistic individuals integrated the two Bayesian components into their decision behavior, and did so indistinguishably from non-autistic individuals. Both groups adjusted their decision criteria in a suboptimal manner. Limitations: This study focuses on explicit priors in a perceptual categorization task and high-functioning adults. Thus, although the findings provide strong evidence against a general and basic alteration in prior integration in autism, they cannot rule out more specific cases of reduced prior effect – such as due to implicit prior learning, particular level of decision making (e.g., social), and level of functioning of the autistic person. Conclusions: These results reveal intact inference for autistic individuals during perceptual decision-making, challenging the notion that Bayesian computations are fundamentally altered in autism.