Active inference is a general framework for optimizing the behavior and learning of a sentient agential system. It may be interpreted as a general theory of sentient behavior and has been used to quantitatively model a wide variety of perceptual and behavioral contexts. Moreover, variables in neural process models of active inference appear to be represented by specific pathways in the brain, and they predict some features of actual neural responses and behavioral patterns in a variety of contexts. These applications support the validity of the active inference framework for describing real animals. However, implementing active inference in a conscious agent requires a system capable of sophisticated probabilistic computations, including a weighted average over its potential future trajectories—a path integral. Although it is straightforward to construct realistic classical biophysical neural models to approximate these computations in simple contexts, we argue in this first of two companion papers that classical Hodgkin-Huxley-style neurons are unlikely to be capable of performing these computations quickly enough in a realistic context. We then explain that conscious (temporally deep) active inference is mathematically equivalent to the path integral that underlies quantum dynamics. A quantum model thus provides a natural, biologically plausible mechanistic implementation of the processing required by active inference. In the second paper we review independent strong theoretical and experimental evidence from my (Wiest) lab and others’ supporting the “Orch OR” quantum theory of consciousness as a collective quantum property of intraneuronal microtubules, which explains the existence of discrete cycles of perceptual inference.

Microtubules

Orch OR

Predictive coding

Active inference

Consciousness

Optimal control

Bayesian brain

© 2025 The Author(s). Published by Elsevier B.V. on behalf of Research Network of Computational and Structural Biotechnology.