Abstract
Visual processing is known to be atypical in Autism Spectrum Disorder (ASD), but it is not understood why. A recent and influential Bayesian account of sensory processing in ASD suggested that the interpretation of sensory input is disrupted in individuals with autism (Pellicano and Burr, 2012) and we have hypothesised that perceptual difficulties in ASD may result from a failure of beliefs (estimated confidence) about beliefs (percepts) that is, formally, a failure of metacognition (Friston, Lawson & Frith., 2013). Thus, visual processing itself may not impaired in ASD, but instead the ability to correctly identify the fidelity of ones' own perception. To test this hypothesis we measured perceptual metacognitive efficiency using a non-social visual discrimination task in in adults with and without ASD. Here our primary goal was to dissociate metacognitive sensitivity from general perceptual performance and response biases (Fleming and Lau, 2014). To do so we used a recently developed and validated (Maniscalco and Lau, 2014) decision theoretic measure of metacognitive efficiency to quantify the extent to which confidence ratings can discriminate between correct and incorrect trials in a perceptual (number discrimination) task. Despite equivalent perceptual performance across multiple measures (fixed at 75% accuracy via staircase), adults with ASD showed suboptimal metacognitive efficiency relative to adults without ASD manifest as a reduced correspondence between trial-by-trial fluctuations in confidence judgements with respect to objective perceptual performance. Additional analyses revealed that adults with ASD showed greater overall uncertainty in their responses (e.g. posterior perceptual beliefs) which would offer some initial support for Bayesian accounts that propose "weak priors" in autism. Our results show for the first time that basic perceptual metacognitive ability is compromised in ASD, offering new mechanistic insights into common perceptual symptoms associated with the condition.
Meeting abstract presented at VSS 2017