The Perceptual Prediction Paradox

From streams of noisy sensory information, we must generate perceptual experiences that are broadly accurate (veridical) and tell us what we did not already know (informative). Bayesian cognitive models propose that predispositions to perceive what we expect will generate more veridical experiences. However, conflicting cancellation models propose that we should perceptually prioritise the unexpected because it is informative. Recent findings from the learning literature may suggest a resolution. We may be initially predisposed to perceive what we expect to generate broadly accurate perceptual representations. However, later processes subsequently highlight unexpected signals that are sufficiently ‘surprising’. We will therefore broadly perceive what we expect, unless unexpected signals are likely to be informative for model updating.