Dynamically speaking, the whole embodied, active system here self-organizes around the organismically-computable quantity “prediction error”. This is what delivers that multi-level, multi-area, grip on the evolving sensory barrage – a grip that must span multiple spatial and temporal scales. Such a grip simultaneously determines perception and action, and it selects (enacts) the ongoing stream of sensory bombardment itself. The generative model that here issues sensory predictions is thus nothing but that multi-level, multi-area[8], multi-scale, body-and-action involving grip on the unfolding sensory stream. To achieve that grip is to know the structured and meaningful world that we encounter in experience and action.
Is this an inner economy bloated with representations, detached from the world? Not at all. This is an inner economy geared for action, whose inner states bear contents in virtue of the way they lock embodied agents onto properties and features of their worlds. But it is simultaneously a structured economy built of nested systems, whose communal project is both to model and engage the (organism-relative) world.