Attention and Explicit Knowledge Drive Predictive Sharpening in Early Visual Cortex

Perception is increasingly understood as an inferential process, whereby what we perceive results from the integration of sensory inputs with expectations derived from prior knowledge. Top-down predictions have been shown to alter the encoding of sensory information, from early to late stages of processing. Yet, how such predictions shape neural representations in sensory cortices remains debated. Competing accounts suggest that predictions either sharpen neural representations by enhancing selectivity or dampen activity by broadly suppressing stimulus-driven responses. In a preregistered fMRI study, we tested whether these effects depend on the level of attentional engagement and explicit knowledge of predictive associations. Using a multisensory fMRI paradigm with concurrent but independent visual and auditory probabilistic associations (75% validity) and manipulated attention, we investigated predictive effects in human early visual cortex. Consistent with prior work, expected visual stimuli elicited reduced BOLD activity. Critically, sharpening of expected visual stimuli occurred exclusively when visual inputs were attended and the concurrently presented auditory inputs expected. In addition, the magnitude of the sharpening of visual representations correlated positively with participants explicit knowledge of the visuo-predictive associations. These findings highlight the key roles of attention and explicit knowledge in promoting predictive sharpening and underscore the need to study predictive processing in more ecologically valid, multisensory contexts.