Learned Visual Relevance Shapes Perceptual Inference: Evidence from Eyes to Hands

Abstract

Perceptual inference requires the brain to decide which signals come from common causes and should hence be integrated or else be segregated. Leading accounts posit that sensory signals are weighted near-optimally by their relative precision consistent with Bayesian principles. It is unknown whether learnt behavioural relevance influences how the brain weighs signals in sensory integration. This study investigated whether visual priority acquired via instrumental learning can reweight sensory inputs in a subsequent unrelated spatial localisation task. In the visual instrumental learning phase, participants responded to one of two differently coloured visual flashes presented in a cluttered environment while withholding their response to the other one. In the audiovisual phase, they located brief sounds presented synchronously midway between brief flashes of the previously relevant target and the competing non-target. Observers' perceived sound location shifted toward the previously relevant target as audiovisual disparity increased. Crucially, this audiovisual spatial bias arose dynamically across the entire cascade of perceptual inference and decision making: from rapid saccades and sustained gaze orienting towards the target location, both occurring after the target's disappearance to initial deviations of manual response trajectories and on a fraction of trials, a change of mind with a return towards the sound as accumulating evidence suggested a separate cause structure. Perceptual inference is thus governed not only by sensory precision but also by visual priorities acquired through instrumental learning. What we learn to prioritise in vision changes where we perceive sounds even in novel and unrelated contexts.