When believing is seeing: the role of predictions in shaping visual perception

The study investigated the influence of predictions on perception-in particular, how strong but erroneous prediction coupled with poor sensory data can trigger misperceptions. Using signal detection, we tested whether predictions induced by a semantic cue change the recognition threshold (criterion) or the subjectively perceived differences between stimuli (sensitivity). In a series of 3 experiments, participants observed pictures of (a) real objects distorted to various extents (targets), (b) noise with elements corresponding to meaningful objects (foils), or (c) nonsense images (noise). Stimuli were preceded either by a semantic cue or by a blank screen with equal probability. Presence of the cue resulted in a more liberal criterion, but only for targets and foils. The cue decreased sensitivity between the distorted targets, but increased sensitivity between the targets and noise-that is, the cue increased between-class differences, but decreased within-class differences. When there was no correspondence between prediction and the sensory input, prediction actually increased the chances of correctly rejecting noise. The influence of the cue was strongest for the moderately distorted targets and foils-when uncertainty related to the bottom-up input was the highest.