Beyond Biology: FEP as a Theory of Everything That Persists

Series: The Free Energy Principle | Part: 7 of 11

Karl Friston didn't set out to explain brains. He set out to explain anything that maintains its structure over time.

Brains are a special case—sophisticated, hierarchical, with exquisite sensitivity to prediction error. But the principle generalizes. Cells minimize free energy through genetic regulatory networks. Immune systems minimize it through clonal selection. Ecosystems might minimize it through trophic dynamics. Even societies could be understood as free energy minimizers at civilization scale.

This is Friston's most audacious claim: the Free Energy Principle isn't biology. It's physics. Any system that persists far from thermodynamic equilibrium—any "thing" that stays organized instead of dissolving into chaos—must be minimizing something equivalent to variational free energy.

If he's right, FEP is a fundamental law of self-organization. If he's wrong, it's still the most ambitious unification attempt in cognitive science.

Let's see how far it stretches.

The Thermodynamic Requirement

Start with the second law of thermodynamics: entropy increases. Ordered states decay into disorder. Structure dissipates. Everything tends toward maximum entropy equilibrium—heat death, sameness, diffusion.

But you exist. You're a highly ordered, low-entropy configuration that maintains itself despite this universal tendency. How?

The classical answer: you're a dissipative structure. You import free energy (food, sunlight), use it to maintain organization, and export entropy (heat, waste). You're open to energy and matter flow, allowing local entropy reduction while increasing global entropy.

Friston adds: the mechanism by which dissipative structures maintain themselves is free energy minimization.

To stay far from equilibrium, a system must have a boundary (Markov blanket) that separates internal from external states. Across that boundary, the system must:

1. Sense the environment (to know what states it's in)
2. Infer hidden causes (to model what's out there)
3. Act (to remain in viable states or move toward them)

This is exactly what free energy minimization does. The FEP describes the informational structure required for persistence, not just the energetic flows.

Cells as Free Energy Minimizers

Take a bacterium. No brain. No nervous system. No representational thought.

But it has:

• Markov blanket: Cell membrane with receptors (sensory states) and pumps (active states)
• Internal states: Metabolic networks, gene expression patterns
• External states: Chemical gradients, temperature, pH
• Free energy minimization: Gene regulatory networks predict viable environmental conditions and metabolic pathways act to maintain them

When a bacterium moves up a glucose gradient, it's minimizing expected free energy. It has (encoded in its biochemistry) a generative model: "when I'm in viable states, I sense high glucose." Encountering low glucose creates prediction error. Chemotaxis is the action that minimizes it.

No consciousness required. No neural tissue. Just molecular dynamics that happen to implement variational inference.

This isn't metaphor—Friston and collaborators have shown that genetic regulatory networks perform approximate Bayesian inference, with mRNA levels encoding beliefs and transcription factors updating them based on environmental signals.

Evolution as Multi-Generational Inference

If organisms minimize free energy over their lifetimes, evolution minimizes free energy over generations.

Natural selection is inference at the population level. Genotypes are hypotheses about viable phenotypes. The environment is data. Reproduction is belief updating—successful genotypes increase in frequency, unsuccessful ones decrease.

Fitness is inverse expected free energy. Organisms with genotypes that predict viable states accurately (and act to fulfill those predictions) survive and reproduce. Those whose predictions fail don't.

Adaptation is model refinement. Over generations, the population-level generative model improves—genes encode better priors about what environments exist and how to persist in them.

This is slow inference (evolutionary timescales) complementing fast inference (lifetime learning). Both minimize free energy. One updates gene frequencies, the other updates neural weights.

Social Systems as Collective Minimizers

Can FEP scale to groups?

A society has:

• Markov blanket: Communication channels (language, media) and collective actions (policy, trade, infrastructure)
• Internal states: Cultural beliefs, institutions, power structures
• External states: Other societies, ecological conditions, resource availability
• Generative model: Shared narratives about what the world is like and how to navigate it

Societies persist by minimizing collective free energy: maintaining coherent identity (low internal surprise) while adapting to changing conditions (epistemic learning).

Institutions are crystallized policies—stable strategies for minimizing expected free energy (courts minimize legal uncertainty, markets minimize resource allocation error, education minimizes epistemic uncertainty in the next generation).

Culture is a shared generative model—agreed-upon priors about what's real, what matters, and how to act. Cultural coherence is low free energy at the society scale.

Collapse happens when the model breaks—when events chronically violate expectations and no policy reduces free energy. Revolutions are catastrophic model updates.

This isn't just analogy. If you can define a Markov blanket and show that internal states evolve to minimize divergence between predictions and observations, you have a free energy minimizer. Scale doesn't matter.

Ecosystems and the Gaia Hypothesis

Can the Earth itself be understood as minimizing free energy?

The Gaia hypothesis (Lovelock) suggests Earth's biosphere acts as a self-regulating system maintaining conditions suitable for life. Temperature, atmospheric composition, ocean chemistry—all held within ranges that allow persistence.

Traditional biology explains this through selection: organisms that stabilize their environment outcompete those that don't. But Friston offers a deeper formalization:

Earth's biosphere has:

• Markov blanket: Atmosphere and oceans (mediating exchange with space)
• Internal states: Biomass distribution, biogeochemical cycles
• External states: Solar radiation, cosmic impacts
• Generative model: Coupled dynamics that predict and maintain conditions within bounds

Gaia minimizes free energy at planetary scale—maintaining states far from equilibrium (oxygen-rich atmosphere, liquid water, stable temperature) by acting (through biological and geological processes) to fulfill predictions about viable planetary conditions.

Controversial? Yes. Testable? Possibly. If you can formalize the relevant Markov blanket and show that planetary dynamics reduce prediction error, it's FEP all the way up.

Consciousness as High-Order Inference

If FEP applies to cells and societies, what about consciousness?

One proposal: consciousness is what it's like to be a high-order generative model.

Simple systems (bacteria, thermostats) minimize free energy without awareness. Complex hierarchical systems (brains) minimize it through layered models that include models of themselves.

Consciousness might emerge when:

1. The system models itself as having a boundary (self/world distinction)
2. The system models itself as modeling (meta-cognition)
3. The system predicts its own future states (temporal continuity of self)

If this is right, consciousness is a particular kind of Markov blanket—one that includes itself in the generative model. A boundary that knows it's a boundary. A process that predicts its own persistence.

No dualism required. No extra ingredient. Just sufficient hierarchical depth and self-referential structure in the inference architecture.

The Universe as Inference?

Friston sometimes ventures further: what if the universe itself minimizes free energy?

The idea: any dynamical system with attracting states (stable configurations it tends toward) can be described as minimizing a free energy functional. Physics tends toward equilibrium. Equilibrium is minimum free energy.

But life goes the other way—maintaining far-from-equilibrium states. How?

By carving boundaries and performing inference. The universe as a whole may tend toward heat death, but local subsystems (with Markov blankets) create pockets of order by acting as if they're minimizing surprise.

FEP becomes a bridge between thermodynamics (systems tend toward high entropy) and biology (systems maintain low entropy). The second law still holds globally. But locally, systems with boundaries can swim against the entropic tide by predicting and acting.

Whether this makes FEP a fundamental law of physics or just a powerful framework for describing life is an open question.

Where FEP Definitely Applies

Even if the most speculative claims don't pan out, FEP clearly applies to:

Cells: Genetic regulatory networks as inference machines
Immune systems: Clonal selection as belief updating
Neural systems: Predictive coding, active inference
Organisms: Homeostasis as surprise minimization
Development: Morphogenesis as prediction-driven self-assembly
Learning: Synaptic plasticity as model refinement
Social groups: Cultural coherence as collective free energy minimization

These aren't analogies. They're systems with definable Markov blankets, generative models, and dynamics that reduce prediction error.

The Radical Implication

If FEP is truly universal, identity becomes function, not substance.

You're not a thing that happens to minimize free energy. You are free energy minimization. The pattern persists because it's the kind of pattern that minimizes its own dissolution.

Cells, organisms, societies, ecosystems—wherever structure persists, there's a Markov blanket and a process functionally equivalent to Bayesian inference.

The universe is full of things trying not to disappear. And the way you try not to disappear is by predicting what states you need to stay in and acting to keep yourself there.

FEP is the math of persistence.

Further Reading

• Friston, K. (2019). "A free energy principle for a particular physics." arXiv preprint arXiv:1906.10184.
• Ramstead, M. J., Badcock, P. B., & Friston, K. J. (2018). "Answering Schrödinger's question: A free-energy formulation." Physics of Life Reviews, 24, 1-16.
• Campbell, J. O. (2016). "Universal Darwinism as a process of Bayesian inference." Frontiers in Systems Neuroscience, 10, 49.
• Rubin, S., Parr, T., Da Costa, L., & Friston, K. (2020). "Future climates: Markov blankets and active inference in the biosphere." Journal of the Royal Society Interface, 17(172), 20200503.

This is Part 7 of the Free Energy Principle series, exploring how FEP might extend beyond brains to all self-organizing systems.

Previous: The Bayesian Brain
Next: Critics and Controversies: What FEP Gets Wrong (Maybe)