Convergence Encyclopedia: C13 — Free Energy / Active Inference

F1 — Tier. T2 (contested — mathematically sophisticated, empirically supported in specific domains, but criticized as potentially unfalsifiable). Uncertainty flag: The Free Energy Principle (FEP) may be able to accommodate any observation post hoc; critics argue this is a feature, not a bug, but it weakens the convergence claim.

F2 — Sources.

• Helmholtz, H. von (1867). Handbuch der Physiologischen Optik. Voss. (Helmholtz free energy in thermodynamics derives from his work on perception.)
• Friston, K. (2005). “A free energy principle for the brain.” Journal of Physiology-Paris, 100(1–3), 70–87.
• Friston, K. (2010). “The free-energy principle: a unified brain theory?” Nature Reviews Neuroscience, 11(2), 127–138.
• Friston, K., Kilner, J. & Harrison, L. (2006). “A free energy principle for the brain.” Journal of Physiology-Paris, 100(1–3), 70–87.
• Rao, R.P.N. & Ballard, D.H. (1999). “Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects.” Nature Neuroscience, 2(1), 79–87.

F3 — Domains. Neuroscience (perception as inference), AI (predictive coding, variational autoencoders), biology (homeostasis as inference), psychology (perceptual inference, action selection).

F4 — Scale. Single neuron (~10⁻⁵ m) → cortical networks (~10⁻² m); formal framework applies at any scale where a system maintains boundaries.

F5 — Falsifier. An adaptive agent that does not reduce prediction error (or variational free energy) yet survives and reproduces — a system that thrives while systematically maximizing surprisal. (Note: critics argue FEP can redescribe any behavior as free-energy minimization, making this falsifier difficult to apply. See rival below.)

F6 — Rival (strongest form). The Free Energy Principle is unfalsifiable — it is a mathematical tautology that any self-organizing system must minimize free energy, because free energy is defined as the difference between the system’s model and the true distribution. Any behavior can be described post hoc as free-energy minimization. This makes FEP a useful modeling framework but not a scientific theory. Critics: Bekesy (2019) Physics of Life Reviews; Clark (2013) Behavioral and Brain Sciences 36(3):181 notes predictive coding is a “substantive empirical hypothesis” while FEP is more ambitious; Colombo & Wright (2018) British Journal for the Philosophy of Science argue FEP lacks empirical content independent of its component models.

F7 — Independence. MODERATE — partial lineage. Helmholtz (19th-century physiology/physics) established the theoretical framework for perception as unconscious inference. Friston (21st-century neuroscience, UCL) developed active inference from statistical physics and Bayesian brain theory. Rao & Ballard (1999) independently developed predictive coding. The lineage from Helmholtz to Friston is conceptual, not institutional.

F8 — Pattern type. Biological / mathematical.

F9 — Maps. A3 (pattern-dynamics), A2 (thermodynamic/computational).

---

Corpus map

• Previous: Convergence Encyclopedia: C12
• Next: Convergence Encyclopedia: C14
• Encyclopedia start: The Schema
• Same node, other planes: Catalogue node C13 · Catalogue hub
• Edges touching C13: disconfirming edge 2
• Kin corpora: Total Structure · Signature of the Grain