Excerpt: The main purpose of this chapter has been to introduce the notion of sensory perception as a form of probabilistic prediction involving a hierarchy of generative models. This broad vision brings together frontline research in machine learning and a growing body of neuroscientific conjecture and evidence. It provides a simple and elegant account of multimodal and crossmodal effects in perception and has implications for the study of (the neural correlates of) conscious experience. It also suggests, or so I have argued, a deep unity between perceiving and imagining. For to perceive the world (at least as we do) is to deploy internal resources capable of endogenously generating those same sensory effects: capable, that is, of generating those same activation patterns via a top-down sweep involving multiple intermediate layers of processing. That suggests a fundamental linkage between ‘passive perception’ and active imagining, with each capacity being continuously bootstrapped by the other. Perceiving and imagining (if these models are on the right track) are simultaneous effects of a single underlying neural strategy.