Humans use overconfident estimates of auditory spatial and temporal uncertainty for perceptual inference

Making decisions based on noisy sensory information is a crucial function of the brain. Various decisions take each sensory signal’s uncertainty into account. Here, we investigated whether perceptual inferences rely on accurate estimates of sensory uncertainty. Participants completed a set of auditory, visual, and audiovisual spatial as well as temporal tasks. We fitted Bayesian observer models of each task to every participant’s complete dataset. Crucially, in some model variants the uncertainty estimates employed for perceptual inferences were independent of the actual uncertainty associated with the sensory signals. Model comparisons and analysis of the best-fitting parameters revealed that, in unimodal and bimodal contexts, participants’ perceptual decisions relied on overconfident estimates of auditory spatial and audiovisual temporal uncertainty. These findings challenge the ubiquitous assumption that human behavior optimally accounts for sensory uncertainty regardless of sensory domain.