BioelectricaAlterspan Labs
Goals and control  ·  Article

Free energy principle and active inference

The free energy principle holds that any system that persists in a changing world will appear to minimize a quantity called variational free energy.1 Active inference is the account of perception and action that follows from this.3 Later authors apply the idea from brains to cells and tissues, as one description of goal-directed behaviour.8

Earliest held
2012, Friston K, Ao P
Most discussed in
Active inference, morphogenesis, and…, 2022
In the library
132 passages in 17 works
Rewritten
2026-10-05
18501875190019251950197520002025Friston K, 20131Friston K, 20132Friston K, Levin M, Sengupta B, Pezzulo…, 20153Friston K, Levin M, Sengupta B, Pezzulo…, 20154Ramstead MJD, Constant A, Badcock PB…, 20195Pio-Lopez L, Kuchling F, Tung A…, 20226Fields C, Levin M, 20227Pio-Lopez L, Bischof J, LaPalme JV…, 20238
01

The existence argument

Friston (2013) described the older motivation for the principle. Systems that do not minimize free energy cannot exist, because the entropy of their sensory states would be unbounded and would increase indefinitely.2 On that reading, biological systems must minimize free energy. Friston called this argument somewhat tautological and turned it around. His claim was that any system that exists will appear to minimize free energy and so engage in active inference.1 He added that this apparently inferential or mindful behaviour is almost inevitable. He offered a heuristic proof based on random dynamical systems, then a synthetic primordial soup in which biological self-organization emerges.1

02

Surprise and thermodynamics

Ramstead and colleagues (2019) described variational free energy as a proxy for surprise, which reflects how improbable a sensory state is for an organism.5 The organism cannot evaluate surprise directly. Its dynamics instead bound a variational quantity that surprise can never exceed.5 Friston, Levin, Sengupta and Pezzulo (2015) said that changes in internal states and in active states both minimize this quantity, giving Bayes-optimal perception and action.3 They also stressed that it is distinct from thermodynamic free energy in physics.4 They left open how the two relate, including whether external states could be cast as a thermal reservoir and internal states as a driven system.4

03

Priors and goal-directed action

Pio-Lopez and colleagues (2022) set out the roles within active inference. Perception selects the predictions that best explain current sensations. Action changes the world so that it yields the agent's preferred sensations, which are encoded in its priors or prior preferences. Learning updates the generative model.6 Fields and Levin (2022) noted an early insight of the approach: minimizing free energy requires probing the environment to see how it responds when perturbed.7 They treated this as productive from cells and tissues up to communities and ecosystems.7 Pio-Lopez and colleagues (2023) cite the framework as integrating homeostasis and allostasis, applied to morphogenetic systems.8

SourcesEach quotation was checked word for word against the passage it opens.
  1. any system that exists will appear to minimize free energy and therefore engage in active inference.Friston K, 2013 · Life as we know it · open at passage 6
  2. systems that do not minimize free energy cannot exist, because the entropy of their sensory states would not be bounded and would increase indefinitelyFriston K, 2013 · Life as we know it · open at passage 6
  3. Both the changes in internal and active states minimize variational free energy, resulting in Bayes-optimal perception and action, respectively. This is known as active inference.Friston K, Levin M, Sengupta B, Pezzulo G, 2015 · Knowing one's place: a free-energy approach to pattern regulation · open at passage 5
  4. The premise underlying active inference is based upon a variational free energy that is distinct from thermodynamic free energy in physics.Friston K, Levin M, Sengupta B, Pezzulo G, 2015 · Knowing one's place: a free-energy approach to pattern regulation · open at passage 37
  5. Technically, variational free energy is a proxy for a quantity called surprise (a.k.a. surprisal), which reflects the improbability of finding an organism in some sensory stateRamstead MJD, Constant A, Badcock PB, Friston KJ, 2019 · Variational ecology and the physics of sentient systems · open at passage 11
  6. action serves to change the world in such a way that it generates the agent's preferred sensations (as encoded in its priors, or prior preferences)Pio-Lopez L, Kuchling F, Tung A, Pezzulo G, Levin M, 2022 · Active inference, morphogenesis, and computational psychiatry · open at passage 6
  7. An early insight of this approach [27] is that minimizing VFE requires probing the environment to determine how it behaves when actively perturbed.Fields C, Levin M, 2022 · Competency in Navigating Arbitrary Spaces as an Invariant for Analyzing… · open at passage 11
  8. both of which have recently been developed to integrate homeostasis and allostasis [97], as applied to morphogenetic systems [98,99].Pio-Lopez L, Bischof J, LaPalme JV, Levin M, 2023 · The scaling of goals from cellular to anatomical homeostasis: an evolutionary… · open at passage 84
Linked ideas
Active inference gives goal-directedness a formal reading: preferred states are encoded as priors, and action brings the world toward them.
Pio-Lopez and colleagues cite the framework as integrating homeostasis and allostasis, with preferred states acting as attractors.
Basal cognitionrelated to
Fields and Levin apply the same formalism from cells and tissues to communities, supporting a cognition-like description below the level of brains.
The framework has been applied to morphogenetic systems, where a target form can be read as a prior preference.
Trial and errorrelated to
Fields and Levin note that minimizing free energy requires probing the environment by perturbing it, which resembles exploratory trial behaviour.
Vitalism and mechanismin tension with
Friston calls the resulting inferential or mindful behaviour almost inevitable for any persisting system, a mechanistic route to what vitalists attributed to special causes.
Friston and Ao derive exploration and exploitation from priors, where bandit heuristics prescribe explicit sampling rules.
Evolvabilityrelated to
Friston 2013 suggests that persisting under the free energy principle implies adaptation and evolution; Levin 2019 and 2023 treat active inference as driving evolved homeostatic loops, and say its role in evolutionary advances needs study.
Niche constructionpart of / contains
The variational approach to niche construction applies the free energy formulation, treating the constructed environment as a resource for active inference.
Ramstead and colleagues apply the free-energy principle across micro and macro scales. Causal emergence offers a different formal way of comparing scales.
The Markov blanket is the formal device of the free energy principle, which derives self-maintenance and apparent inference from it.
Passages tie them closely: Fields & Levin cast variational free energy as prediction-error minimisation, and Friston frames free-energy minimisation as action and perception. Allen & Friston treat predictive processing as a broad family.
Gradient descent on free energy is the formal setting in which Friston and Ao, and later Pio-Lopez and colleagues, frame flow in state space.
The framework in which Friston introduced blankets as the statistical basis of lifelike behaviour.
Allen and Friston treat organism, body-brain and world as one generative model, another way of tying environment to the observer.
Supplies the variational and posterior vocabulary that Kirchhoff and colleagues use to frame inference as minimizing free energy.
Hierarchical dynamic models and variational energy supply the mathematical machinery in which the free-energy principle and active inference are expressed.
Fields and Levin use variational free energy minimization as an invariant meant to hold across diverse embodiments.
Precision is a parameter within the free-energy formulation, where it weights prediction errors against priors.
The affordance account is built on active inference, which supplies the generative model in which cues, precision and action are inferred.
Ramstead and colleagues describe organisms as occupying a bounded viability set within phase space, which links this formalism to attractor language.
Sengupta and colleagues suggest free-energy optimisation might explain synaptic pruning during neurodevelopment.
Where it is discussedPassages matching free energy principle, active inference, variational free energy, scalar cost function
2022Active inference, morphogenesis, and computational psychiatry · Pio-Lopez L, Kuchling F, Tung A…38
2019Variational ecology and the physics of sentient systems · Ramstead MJD, Constant A, Badcock PB…29
2015Knowing one's place: a free-energy approach to pattern regulation · Friston K, Levin M, Sengupta B, Pezzulo…16
2012Free energy, value, and attractors · Friston K, Ao P11
2022Competency in Navigating Arbitrary Spaces as an Invariant for Analyzing… · Fields C, Levin M10
2013Life as we know it · Friston K9
2025The Multiscale Wisdom of the Body: Collective Intelligence as a Tractable… · Levin M3
2022Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework… · Levin M3
2023The scaling of goals from cellular to anatomical homeostasis: an evolutionary… · Pio-Lopez L, Bischof J, LaPalme JV…2
2024Collective intelligence: A unifying concept for integrating biology across… · McMillen P, Levin M2