Free Energy Principle (Karl Friston)

Core Results

Karl Friston proposed that self-organizing systems minimize variational free energy. This bound on surprise maintains low entropy states. Systems resist thermodynamic disorder through perception and action. The principle unifies action, perception, and learning under one imperative.

Biological agents maintain a small repertoire of physiological states. They do so by minimizing the long-term average of surprise. Free energy provides an upper bound on this surprise that agents can evaluate from internal states and sensory data.

Action samples data consistent with current predictions. Perception updates internal models to reduce prediction error. The result is approximate Bayesian inference in self-organizing systems.

Primary Works

Friston, K. The free-energy principle: a unified brain theory?. Nat Rev Neurosci. 2010 Feb;11(2):127-38. doi: 10.1038/nrn2787.

Quote: "A recently proposed free-energy principle for adaptive systems tries to provide a unified account of action, perception and learning."

Quote: "The free-energy principle says that any self-organizing system that is at equilibrium with its environment must minimize its free energy."

Earlier notes appear in Friston K, Kilner J, Harrison L. A free energy principle for the brain. J Physiol Paris. 2006;100(1-3):70-87.

Later extensions cover active inference and consciousness. Friston K. Sentience and the Origins of Consciousness. Entropy. 2020;22(5):516.

Convergence Patterns

The principle derives energy flow to structure. Thermodynamic free-energy minimization produces stable attractors. These attractors correspond to bounded states across scales.

It derives structure to memory and inference. Generative models encode expectations. Prediction error drives updates that build internal representations.

It derives inference to mind. Active inference couples perception and action. This coupling grounds basic forms of sentience under Markov blankets.

The work touches the Ladder at the step from flow to structure and from structure to memory. It reaches inference and basic consciousness. See /a/oip-the-ladder.

Distance from the Full Synthesis

Friston stops at self-organizing systems and brains. The synthesis extends the grain to universal patterns such as branching, spirals, waves, and scale invariance in non-living flows. FEP supplies the thermodynamic engine but does not enumerate those geometric outcomes.

FEP places the observer inside the system via Markov blankets. This aligns with the Mirror Layer. The synthesis adds explicit reader-system recursion across all scales. See /a/oip-the-mirror-layer.

FEP derives mind from free-energy minimization. The synthesis adds the full sequence through life and explicit memory mechanisms. FEP reaches mind but does not detail the intermediate biological steps in equal depth.

Limits and Objections

FEP rests on formal mathematics of variational inference. It has limited direct empirical tests at the organism level. Most support remains mechanistic.

A key internal objection questions the metaphysical use of Markov blankets. Bruineberg J, et al. The Emperor's New Markov Blankets. Behav Brain Sci. 2022;45:e200. doi:10.1017/S0140525X21002275.

The critique distinguishes epistemic blankets from physical boundaries. It argues that claims about agent-environment demarcation overstep the original formal tool.

FEP does not address whether minimization produces the full family of grain patterns outside biology. Reductionist accounts in the style of Weinberg treat the principle as one layer among many without universal reach.

Claims

• Claim c1: Self-organizing systems minimize variational free energy to resist disorder. Tier: mechanistic. Section: Core Results. Source: s1.
• Claim c2: Free energy bounds surprise and enables approximate Bayesian inference. Tier: mechanistic. Section: Core Results. Source: s1.
• Claim c3: Action and perception jointly minimize free energy under generative models. Tier: mechanistic. Section: Core Results. Source: s1.
• Claim c4: The 2010 paper states that any self-organizing system at equilibrium minimizes free energy. Tier: anecdotal. Section: Primary Works. Source: s1.
• Claim c5: FEP derives stable attractors from thermodynamic imperatives. Tier: mechanistic. Section: Convergence Patterns. Source: s1.
• Claim c6: FEP reaches inference and basic sentience but omits explicit grain geometries such as spirals. Tier: speculative. Section: Distance from the Full Synthesis. Source: s2.
• Claim c7: Markov blankets demarcate internal states in FEP. Tier: mechanistic. Section: Distance from the Full Synthesis. Source: s1.
• Claim c8: Bruineberg 2022 critiques the metaphysical extension of Markov blankets. Tier: anecdotal. Section: Limits and Objections. Source: s2.