Predictive Processing and Active Inference: A Comprehensive Review of Theoretical Foundations, Neural Mechanisms, and Clinical Implications in Cognitive Science

Abstract

Predictive processing (PP) and its active counterpart, active inference (AI), have emerged as among the most influential frameworks in contemporary cognitive science, offering a unified account of perception, cognition, action, and learning under a single computational principle: the minimization of prediction error or, more formally, the minimization of variational free energy. Originally grounded in Helmholtzian notions of perception as unconscious inference, the framework has been substantially formalized through Karl Friston's free energy principle (FEP), which proposes that all self-organizing biological systems resist entropy by implicitly minimizing surprise. This review systematically examines the theoretical underpinnings of the PP/AI framework, its neurobiological implementation via hierarchical predictive coding in cortical circuits, and its behavioral and clinical manifestations across a spectrum of psychological conditions. We synthesize evidence from computational modeling, electrophysiology, neuroimaging, and psychophysics to evaluate the empirical status of the framework. We further discuss unresolved controversies—including the explanatory scope of the FEP, the interpretation of precision-weighting, and the relationship between PP and alternative computational theories of mind—and identify promising directions for future research. We conclude that while PP/AI represents a transformative theoretical advance, its full explanatory power remains contingent on tighter integration with mechanistic neuroscience and rigorous empirical testing.