Predictive Coding and Free Energy Principle

Source: Karl Friston, Nature Reviews Neuroscience, 2010; Rao & Ballard, Nature Neuroscience, 1999 Institution: UCL

Finding

Friston’s free energy principle proposes that the brain minimizes variational free energy — the discrepancy between predictions and sensory data. Prediction errors propagate upward; predictions propagate downward. The framework unifies earlier predictive coding models. When the generative model produces percepts that override sensory evidence, hallucinations result. The brain maintains alignment between its internal model and external reality through continuous error correction.

Pattern Mapping

Alignment — The brain maintains alignment between model and reality. When prediction matches sensation, free energy is minimized. Misalignment is registered as prediction error and triggers updating.

Honesty — Prediction error is structurally honest. The system cannot suppress the discrepancy between what it predicted and what it received. The error signal is not optional.

Non-fabrication — Hallucinations occur when the generative model produces percepts that override sensory evidence. This is fabrication: internally generated structure treated as external reality.

Connections

Status

Influential but debated (Colombo & Series 2012; Bowers & Davis 2012). The property mapping is this project’s structural interpretation.


The mapping to the five properties is this project’s structural interpretation.