Predictive processing for sensory and action neural representations

Abstract

Prior information, such as predictive cues, shape perceptual decisions. Predictive theories make conflicting claims about whether the brain preferentially processes expected, as proposed by Bayesian accounts, or unexpected information, as proposed by cancellation accounts. These conflicting mechanisms may co-exist if they differ across sensory and action systems, or if their influence changes as the decision unfolds. To test these alternatives, we tracked sensory and action representations in EEG during a motion-discrimination task with changing cue validity. Cross-temporal decoding revealed distinct stabilization of sensory and action representations. Under strong priors, stabilization of unexpected actions predicted stronger bias by the cue, consistent with cancellation accounts. Unlike action, stabilization of expected sensory information was both positively and negatively correlated with precision-weighted prior bias, depending on the decision stage (before or after stimulus onset), compatible with both Bayesian and cancellation accounts. Together, these findings suggest that predictive processing does not uniformly favor expected or unexpected information. Instead, its effects depend on representational domain and decision stage.