Self-evidencing agents

Jakob Hohwy (Monash University)

Abstract: If we cast predictive processing as self-evidencing rather than just predictive coding, then we can understand perception, belief updating and decision-making as abductive processes. Self-evidencing agents close the action-perception loop in a fundamentally self-supervised way that tracks abduction and unifies normative and descriptive approaches to behaviour. This explains our diverse cognitive trajectories and our often non-linear engagement with ourselves, each other and our environment. I introduce this notion of self-evidencing agents, and propose ways in which it can shift our conceptions of self and volition.