Updating implicit contextual priors with explicit learning for the prediction of social and physical events

For proper action understanding we can infer action intention from the kinematic features of movement (the event in terms of sensorial evidence) and/or from the contextual scenario in which the action occurs. In line with predictive coding theories, the implicit learning of statistical regularities between events and contextual cues strongly biases action prediction. Here, we assessed the relative sensitivity of contextual priors to an explicit learning aimed at reinforcing either context-based or event-based prediction. First, in an implicit learning phase we exposed participants to videos showing specific associations between a contextual cue and a particular event (action or shape) in order to create high or low contextual priors. Then, in an explicit learning phase we provided a feedback reinforcing the response suggested by the contextual prior or by the sensory evidence. We found that the former improved the ability to predict the unfolding of social or physical events embedded in high-probability contexts and worsened the prediction of those embedded in low-probability contexts. Conversely, the latter had weaker effects, ultimately failing to override the reliance on contextual priors. Further, we acknowledged an association between the extent of individual autistic traits and the ability to leverage explicit learning mechanisms encouraging perceptual information.