The role of beliefs about perception in perceptual inference

According to Bayesian, “inverse optics” accounts of vision, perceiving is inferring the most likely state of the world given noisy sensory data. This inference critically depends on an internal model specifying how world states translate to visual sensations. Alternative accounts explain perceptual decisions as a rule-based process, with no role for beliefs about perception itself. Here, we contrast the two alternatives, focusing on decisions about perceptual absence as a critical test case. We present data from three pre-registered experiments where participants performed a near-threshold detection task under different levels of partial stimulus occlusion, thereby visibly manipulating the perceptual likelihood function going from external world states to internal perceptual states. We observe differential effects of sensory evidence and occlusion on decisions about presence and absence. Our normative model of visual detection accounts for these asymmetries, and further reveals robust individual differences, with some participants systematically incorporating beliefs about perception into perceptual inferences more than others. We discuss implications for the varied and inferential nature of visual perception more broadly.