Late Event Related Potentials as prediction error signals that determine the updating of prior information with sensory evidence

Recent proposals argue that our understanding of our dynamic environment emerges from a predictive process that integrates prior expectations with the sensory evidence, implementing a process of Bayesian-like hypothesis testing and revision. In two preregistered EEG/ERP experiments, we investigated the neuronal underpinnings of this integration. Participants observed hand motions towards and away from objects after hearing actors state the intention to "take" or "leave" those objects. The hands disappeared mid-motion and participants reported their perceived last locations on a touch screen. As expected, and consistent with a Bayesian-like integration process, we found, first, that participants’ reports perceptually overshot the observed motions, and these biases were larger when the motions matched the previously stated action intention, particularly when the intentions reliably predicted the motions. Second, in both experiments, the amplitude of the post-stimulus P2 and P3b ERP components closely tracked the integration of observed motion with prior expectations. P2 and P3b were larger when observed motions and prior expectation mismatched, linking these components to the hypothesized mechanism for revising prior expectations in light of conflicting sensory evidence. Third, trial-by-trial modelling confirmed that both the P2 and P3b underpin the expectation revision in light of conflicting sensory evidence, reducing the perceptual overestimation towards the expected next steps on the motion sequence. Our results therefore reveal for the first time that these EEG components reflect the prediction error process, which revises prior expectations towards the sensory evidence, so that these expectations have a reduced influence on perceptual judgments. These findings argue for a framework in which prior expectations – and their revision through stimulus processing – shape perception.