Free Energy Principle

The Free Energy Principle (FEP) proposes that all living organisms maintain their existence by minimizing variational free energyan upper bound on surprisal, defined as the negative log-probability of observed states. Developed primarily by Karl Friston, with contributions from Maxwell Ramstead, Wanja Wiese, Alex Kiefer, and others, FEP describes organisms as performing approximate Bayesian inference, dynamically updating their beliefs to predict and minimize sensory states. The framework applies to self-organizing systems that exhibit a statistical separation between internal, external, and intermediate states through structures called Markov blankets.

The core mechanism involves hierarchical predictive coding, where neural systems generate predictions about sensory inputs and update these predictions based on prediction errors. Consciousness within FEP is framed as the capacity to engage in deep generative modeling and counterfactual reasoning through hierarchical Bayesian inference. The theory links conscious experience to neuromodulation and precision-weighting of predictionsthe process by which the brain determines which prediction errors should most strongly influence belief updating. Attention, in this framework, is understood as precision-weighting that modulates the influence of different signals.

Empirical support for FEP comes from predictive coding models that explain how neural systems update predictions and resolve prediction errors, neuroanatomical evidence that brain structures can be modeled as nested Markov blankets, and physiological studies linking consciousness to precision-weighting mechanisms. The theory's scope extends beyond consciousness to address questions of life, agency, and self-organization more broadly. Critics note that FEP's generality may undermine its falsifiability, and questions remain about substrate-dependence and whether the framework adequately addresses the hard problem of consciousness.