Affordance and Active Inference

Abstract

This chapter considers affordance from the point of view of active inference, namely, a first principle account of how we choose what to sample from our sensorium. In brief, it considers the imperatives for an enactive engagement with the sensed world from the Bayesian perspective of self-evidencing. Put simply, this means perception and action—and everything in between—can be read as maximising the evidence for our internal world (i.e. generative) models that we use to explain the causes of our sensations. Crucially, if actions are considered a cause of sensations, then our behaviour itself has to be inferred (this is sometimes known as planning as inference). This kind of planning can be very elemental (e.g. where should I look next?) or more deliberative (e.g. what kind of policy should I adopt in this situation?). Active inference suggests that policies or plans should be chosen to maximise something called expected free energy. This effectively scores plausible courses of action in terms of their epistemic and pragmatic affordance. The epistemic affordance corresponds to the expected information gain, while pragmatic affordance corresponds to the expected log likelihood of preferred outcomes. In summary, on this view, affordances become an attribute of a certain course of action that depends on both the inferred states of the world and (Bayesian) beliefs about those states of affairs.