thing implemented by System B – is therefore a form of active inference, the process by which sensory prediction errors are minimised through making actions rather than by updating predictions. As I explained in chapter 5, active inference depends both on generative models which are able to predict how the causes of sensory signals respond to different actions, and on modulating the balance between top-down predictions and bottom-up prediction errors, so that perceptual predictions can become self-fulfilling.
