active inference

Compiled with AI assistance and selectively reviewed. Please consult the cited passages; the books remain the authoritative source. About this companion

Editorial synthesis · provisional

Definition

Active inference is Friston’s account, presented in RB2, of action reducing discrepancies between expected and actual sensory states alongside perceptual updating and learning. Organisms can change what they encounter rather than only revise a model. The entry preserves this theoretical account without identifying all action with conscious planning or treating formal descriptions of persistence as proof that every persistent system literally infers.

Explanation

RB2’s proprioceptive example describes anticipated arm position and bodily adjustment, while its sampling examples include moving eyes, head and attention. The defining contrast is between merely receiving input and participating in its selection. It does not require an explicit sentence about a future state before each movement.

The account joins action, perception and learning through the reduction of uncertainty and surprise. Precision gives some discrepancies more weight than others, and material or cultural scaffolds can organize the stream of evidence. These relationships preserve the framework’s embodied ambitions without equating every scaffold with one internal neural representation.

The thermodynamic discussion explicitly separates variational from thermodynamic free energy and records foundational criticisms of the free-energy principle. Modeling a system as if it inferred does not establish literal inference or an identity between life and Bayesian computation. Active inference therefore remains an attributed framework, not a universal proven ontology.

Scope and attribution

Preserve Friston attribution and explicit formal and ontological qualifications.

AB — The Actual Book

Source-verified occurrence; see the printed-page references below.

Occurrence locators: 99.

RB2 — The Real Book, Volume II: Becoming Is Being

Action shapes input within a formal framework whose universal interpretation is qualified.

Occurrence locators: 107, 108, 115, 116, 117, 119, 121, 122, 134.