ABSTRACT
We formulate predictions about the sensory outcomes of our actions. These predictions serve a number of cognitive functions. First, they allow us to select appropriate actions based on their anticipated sensory consequences. For example, we select particular motor commands that will enable us to pick up a cup to drink. Second, these predictions can render our perceptual experiences during action more veridical. The sensory input during action can be especially noisy and change rapidly and if we bias our perception towards what was expected (a hand grasping a cup), it will tend to be more accurate. Third and finally, predictions play an important role in model updating and action control. If our hand did not reach the cup, then we may need to update our models about action-perception correspondences and perform corrective actions. We will discuss a range of models concerning how predictions underlie these distinct cognitive functions, as well as considering how models optimising one function frequently ignore the others. We will end by examining how our recent empirical and theoretical work may resolve these discrepancies, proposing a framework to enable us to use sensory predictions for a range of cognitive functions during action.
