Predictive processing's flirt with transcendental idealism

Abstract

The popular predictive processing (PP) framework posits prediction error minimization (PEM) as the sole mechanism in the brain that can account for all mental phenomena, including consciousness. I first highlight three ambitions associated with major presentations of PP: (1) Completeness (PP aims for a comprehensive account of mental phenomena), (2) Bayesian realism (PP claims that PEM is implemented in the brain rather than providing only a model), and (3) Naturalism (PP is typically presented as yielding a naturalistic view of the mind). Then I demonstrate that many proponents of PP also endorse a form of Kantian transcendental idealism (TI), based on a characterization of experiential content as the brain's currently best hypothesis about the world. I argue that endorsing this claim (4), that is, that we only experience the world as it appears, but not the world itself, sabotages achieving the three ambitions. The argument proceeds by discussing the prospects of each ambition in turn, drawing on discussions in the philosophy of science about realism and its alternatives, about the motivation and features of computational models, and about the foundational role of consciousness for science.