When seeing is not believing: A mechanistic basis for predictive divergence

Visual illusions provide a compelling case for the idea that perception and belief may remain incongruent. This can be explained by modular theories of mind, but it is not straightforwardly accommodated by the Predictive Processing framework, which takes perceptual and cognitive predictions to derive from the same underlying inferential hierarchy. Recent insights concerning the neural implementation of Predictive Processing may help elucidate this. Specifically, prior information is proposed to be approximated by mechanisms in both the top-down and bottom-up streams of information processing. While the former is context-dependent and flexible in updating, the latter is context-independent and difficult to revise. We propose that a stable divergence between perception and belief may emerge when flexible prior information at higher hierarchical levels contradicts inflexible prior information at lower ones. This allows Predictive Processing to account for conflicting percepts and beliefs while still maintaining a hierarchical and unitary conception of cognition.