Predictive Processing without Prediction

Predictive Coding and Predictive Processing are among the most influential theoretical approaches in contemporary cognitive neuroscience. Much of their appeal rests on the fact that neural dynamics are described as a continuous process of prediction, prediction error, and correction. The present article questions neither the mathematical modelability of such dynamics nor the empirical value of locally bounded Predictive Coding models. It asks instead what theoretical content the predictive interpretation itself possesses.

For elementary Predictive Coding models, central mathematical relations can be formulated in mechanistically neutral terms without changing their dynamics. Predictions become endogenously generated neural activities, prediction errors become difference signals, and inference becomes state change due to recurrent coupling. For more comprehensive Predictive Processing and Free Energy models, the situation is different. Their probabilistic and Bayesian mathematics cannot simply be replaced by different terminology. Yet the possibility of modeling biological processes probabilistically or in Bayesian terms does not imply that the modeled system itself generates predictions, possesses beliefs, or performs inference.

Semantic phenomena are not thereby denied. They are kept categorically distinct from physiological description. At the physiological level, neural processes are described in terms of activity, coupling, difference, feedback, state change, and plasticity. Meaning belongs to another description of the same organismic process. It therefore need not already be inscribed into the neural mechanism as prediction, error, belief, or preference.

This yields a dilemma in the philosophy of science. If the central concepts of Predictive Processing possess strong cognitive and semantic content, they require additional empirical justification. If they merely designate formal or dynamical relations, they are theoretically dispensable relative to a mechanistically neutral description.