Abstract
When processing sensory information, our brain relies on predictions to facilitate and accelerate making sense of these. Violating these predictions in an experimental setting is a useful tool to investigate how our brain processes sensory data. However, most studies investigating prediction error signals generally use unlikely, but never (seemingly) impossible events.
In this fMRI study we used videos of magic tricks as naturalistic stimuli that violate predictions based on our intuitive understanding of how the physical world works. Using univariate as well as multivariate analyses we discovered a hierarchy of prediction error signals, with frontoparietal areas responding generically and posterior visual areas responding specifically to violations of object- and feature constancy. These prediction error signals were modulated by prior knowledge, showing that high level contextual predictions are already present in early cortical visual areas.
