Perceptual meta-uncertainty reveals enhanced calibration of sensory confidence in autism.

Atypical metacognition has been suggested to underlie autistic phenotypes, given its role in social cognition and behavioral flexibility. However, no study has quantitatively assessed metacognitive abilities in autism. Here, we measured meta-uncertainty-the noise corrupting estimates of one's own decision uncertainty-in autism. In three experiments, autistic and non-autistic participants (N = 145) performed orientation categorization tasks while simultaneously reporting their choice confidence. By independently manipulating each Bayesian component-sensory uncertainty, prior, and reward-and fitting a recently established process model, we assessed metacognitive abilities and their contingency on the Bayesian components while controlling for first-order decisions. Unlike non-autistic participants, autistic participants' meta-uncertainty depended on which decision component was manipulated and was lower than that of non-autistic participants, specifically when decisions were adjusted for sensory uncertainty. These findings reveal that metacognition in autism is not generally reduced but rather enhanced for inferences that rely primarily on sensory information.