It has long been appreciated that learning about the probabilistic structure of our sensorimotor environment shapes perception in important ways. However, it is less clear how. For example, we must balance demands of representing the perceptual world accurately while effectively using perception to update our models when the world changes.

Bayesian accounts propose that predictions are integrated with sensory inputs to determine what we perceive. Because the prediction has been a source of information that feeds into the percept, we are more likely to perceive what we expect and we perceive it more intensely. Accounts propose that neurally this is achieved by amplifying the sensory gain of expected channels. This mechanism would render our percepts more accurate in a noisy sensory world, because what we predicted to be there is more likely to be there. We can also generate these best guesses more rapidly than if relying solely on sensory input, given the neural transmission delays associated with its processing.  In contrast, attenuation accounts propose the opposite. They claim that we reduce the sensory gain of expected channels, which reduces perception of the expected – rendering us less likely to perceive it and perceiving it less intensely. These accounts highlight a mechanism that would render our percepts more informative in a limited capacity processing system – directing resources to events that tell us more and thereby likely require model-updating. These accounts both seem sensible and are supported by a host of data but are mutually incompatible.

I will ask how we overcome the challenge of optimising sensorimotor perception such that it is accurate, rapid and informative, with accounts that can serve all functions rather than optimising for one at the direct cost to another. I will present work that tests these possibilities by asking what we perceive alongside neural processing across time (EEG) and space (7T fMRI). I hope to convince the audience that our models of action-perception interdependences should move on from some currently-popular monolithic accounts and stimulate discussion concerning how best to conceptualise these synergistic relationships.