Free energy principle
Explanation
Karl Friston, British neuroscientist, has been developing since the 2000s one of the most ambitious and mathematically formalised theoretical proposals in cognitive science: the free energy principle. According to this principle, every adaptive system —from a bacterium to a human brain— can be understood as an agent that minimises a quantity called variational free energy, which is roughly equivalent to surprise or prediction error.
The idea comes from statistical physics and Bayesian machine learning. A system that persists in an environment must remain within certain characteristic states (temperature, pH, glucose, etc., in a biological organism). To do so, it must be able to predict the consequences of its actions and select those that bring it to those states. Minimising free energy is equivalent to reducing the discrepancy between what the system expects and what it senses.
There are two strategies for minimising free energy: updating internal models (perceptual learning) or acting on the world (action) to make predictions come true. Both form part of a single integrated loop. The same framework explains perception, action, learning, attention, emotion: all are forms of minimising surprise.
Active inference is the behavioural version of the principle. Animals act on the world to reduce uncertainty about their states: they explore, gather information, test hypotheses. Childhood curiosity, scientific research, even play, would be manifestations of the same thermodynamic-cognitive imperative to minimise free energy.
For consciousness, the principle suggests a unified perspective: conscious experience would be the global predictive model of the integrated organism, including its own bodily state. Consciousness would emerge when the model becomes self-referential and hierarchically integrated, with meta-predictions about its own predictions. Anil Seth and Jakob Hohwy have developed this line by linking it to predictive processing.
The principle is mathematically sophisticated but remains debated. Some consider it a great unification in the manner of Newton; others, a tautology dressed up in formalism. Empirically testable predictions are hard to derive. But its ambition and scope —linking physics, biology, cognition and perhaps consciousness— make it one of the most discussed proposals in contemporary theoretical neuroscience.
Strengths
- Mathematically formal and unifying framework.
- Connects thermodynamics, information and cognition.
- Clinical and AI applications.
- Compatible with autopoiesis and enactivism.
Main critiques
- Charged with non-falsifiability due to excessive generality.
- Does not explicitly address qualia.
- Mathematical notation hard to relate to direct empirical observables.
- Some assumptions (Markov blankets) are contested.