87 | Karl Friston on Brains, Predictions, and Free Energy
What this covers
Karl Friston and Sean Carroll build a unified theory of how the brain works and, by extension, how any persistent living system maintains itself. The argument rests on a single mechanism: all sentient behavior arises from systems minimizing variational free energy—a mathematical stand-in for surprise—against an internal generative model of the world. Friston connects this to thermodynamic first principles: variational free energy is formally identical to the classical free energy equation from statistical physics, differing only in dropping Boltzmann's constant and operating on beliefs rather than physical states. The conversation moves from the mathematical foundations through neural mechanisms, then turns to psychiatry, where Friston argues that mental illness is fundamentally false inference.
The discussion covers how Markov Blankets—the boundary separating internal from external states—license the existence of autonomous systems, why Bayesian approximation is biologically necessary when exact inference is intractable, and how predictive coding explains belief updating. A major thread concerns precision—the confidence or error bars attached to sensory evidence and beliefs—and its breakdown in psychopathology. Friston argues that nearly all psychiatric conditions reduce to failures in estimating uncertainty, and that most psychiatric drugs target neuromodulators that adjust precision rather than beliefs themselves. The conversation also explores action through active inference, the mechanism by which planning systems imagine future states and act to fulfill them, explaining both voluntary movement and intentionality in ways that distinguish living systems from passive physical structures like hurricanes. Throughout, Friston situates these ideas as practical necessities: understanding and treating patients requires the ability to simulate sentient behavior, including its failures.
Friston argues that all sentient behavior—from thermostats to humans—can be derived from a single first principle: systems persist by minimizing variational free energy (surprise) relative to a generative model of the world, and that psychiatric disorders are best understood as failures of this inferential process, especially failures to encode precision (error bars).
- Variational free energy is mathematically identical to thermodynamic free energy minus Boltzmann's constant, but operates on beliefs rather than physical states
- Systems that exist must possess a Markov blanket separating internal from external states, which implies an internal generative model
- Psychopathology is false inference, and much of it reduces to broken precision (error-bar) estimation, pointing to neuromodulatory drug targets
Cells possess rich internal models of the world unlike self-organizing physical structures.
- Self-organizing physical structures like hurricanes and candle flames do not possess genuine Markov blankets and are not alive in the relevant sense; unlike even the simplest cell, they are not rich with information and carry no model of the world—they merely persist non-ergodically without enduring an attracting set.
“I've been asked does Gaia. No, it doesn't. That's very irritating. Nor do candle flames.”
The brain minimizes surprise by balancing commitment to likely explanations with openness to alternatives.
- You cannot minimize surprise by committing 100% probability to one prediction (any deviation produces huge surprise) nor by spreading belief over everything (then everything is always somewhat surprising); the brain instead homes in on the most likely explanation while keeping options open, which is Occam's Razor and explains why systems occupy a structured but non-singular region of state space.
“you don't want to commit to a very precise one, so you've got to find the simplest, the most accommodating explanation for your data. So that's the paradox”
Non-equilibrium statistical mechanics of dissipative self-organizing systems remains a pre-paradigmatic frontier.
- Non-equilibrium statistical mechanics of open, dissipative, self-organizing systems is a booming but pre-paradigmatic field where the rules, equations, and even the right words are not yet known, representing the frontier of 21st-century physics with major discoveries still to be made.
“physics has a lot of work to do, and there's a lot of discoveries to be made along exactly these lines. Where non-equilibrium statistical mechanics is a booming growth field”
Homeostasis—maintaining the organism within a viable range—centers understanding of life and mind.
- Antonio Damasio centers his understanding of life and mind on homeostasis—keeping the organism within a tiny viable range with some flexibility.
“I did have Antonio Damasio on the podcast a little while and his favorite word in the universe is homeostasis, right, the idea of keeping within this tiny little range as much as we can, but some flexibility there.”