The Larger World ArchiveClaims · Sources · Evidence
Mind, belief and memoryLive debateTheoretical neuroscience · 2006 – present

The Free Energy Principle

A single mathematical principle is said to explain perception, action, learning and life itself. Its defenders call it the closest thing neuroscience has to a unifying law; its critics call it unfalsifiable, and both have a point.

Cortical cells forming dense interconnected networks in vitro. The free energy principle is a claim about what
Cortical cells forming dense interconnected networks in vitro. The free energy principle is a claim about what any system doing this is up to. Image: Authors of the study: Brett J. Kagan Andy C. Kitchen Nhi T. Tran Forough Habibol · CC BY 4.0 — Wikimedia Commons.

The claim

Karl Friston's free energy principle starts from a fact about persistence. Anything that continues to exist must, statistically, keep itself within a limited set of states — a body stays at a temperature, a cell keeps its ion balance, an organism stays alive. The principle formalises this as minimising variational free energy, a quantity from statistical mechanics that bounds surprise.

The Markov blanket: the boundary that separates a system from its environment in the formalism.
Everything inside the blanket is the system; everything outside is the world; the blanket is the interface. Figure: The Larger World Archive.

From that, the framework derives perception as inference about the causes of sensory input, action as changing the world to fit the prediction, and learning as updating the model. Predictive processing is the psychological cousin; active inference is the engineering one.

Why it is attractive

It is one formalism for many phenomena. The same equation is claimed to cover vision, motor control, dopamine signalling, psychosis as aberrant precision, and the behaviour of single cells. It also connects to established mathematics — variational Bayes, the evidence lower bound, control theory — rather than inventing notation.

The objection that sticks

A principle that everything living must satisfy cannot distinguish between hypotheses about how a particular brain works. Critics argue the framework is close to a tautology: if persistence requires staying in a low-surprise region, then systems that persist minimise surprise, and the theory is true by construction. Friston's reply is that the interesting content is in the specific generative models, not in the principle — but those models are usually built by hand for each experiment, which is where the falsifiable work actually happens.

The practical test is whether it has produced findings that would not have been found otherwise. Active inference has produced useful engineering results and some striking neuroscience predictions; it has not produced a decisive experiment that killed a rival.

Where the label sits

Not fringe: the mathematics is mainstream, published in Nature Reviews Neuroscience, and used in real experiments.

Not settled: the claim that it is the unifying principle of brain function is contested by people who work on the same problems.

The useful distinction: predictive processing as a family of models is well supported. The stronger claim that all biological self-organisation is free-energy minimisation is a metaphysical bet with a mathematical costume.

Why it is here

Because it is what a grand unified theory looks like from the inside: elegant, generative, and impossible to falsify at the level where it is stated most broadly. The archive treats it the same way it treats the multiverse — serious, and not yet evidence.

Sources for this page

  • Friston (2010), Nature Reviews Neuroscience — the founding paperdoi.org
  • Wikipedia, “Free energy principle” — formalism and criticismen.wikipedia.org
  • Wikipedia, “Predictive coding” — the psychological cousinen.wikipedia.org
  • Stanford Encyclopedia of Philosophy, “Bayesian models of the mind”plato.stanford.edu