Karl Friston built the Free Energy Principle and active inference, the formal mathematics of how living systems perceive, predict and act on their world under uncertainty. Mark McGrath and I walked him through the sketch of the OODA loop that John Boyd actually drew.
Karl is a Fellow of the Royal Society and a neuroscientist at University College London. I found his work through Michael Pollan, Robin Carhart-Harris and the REBUS model. This is the conversation we have been building toward for 163 episodes.
We build the framework from first principles. Karl starts with the features of the environment Boyd treated as inescapable: entropy, ambiguity and novelty. He describes the Markov blanket, the boundary that defines a self. He then describes the generative model, the internal beliefs a living system uses to infer the hidden causes of what it senses. Boyd called that generative model orientation, and it sits at the center of his real OODA loop. Karl also takes up the two pathways Boyd drew from orientation, implicit guidance and control, which few people have ever explained.
Risk and expected free energy
Karl puts formal language under something Wall Street already senses. Risk is a KL divergence between the outcomes you prefer and the outcomes you predict. A KL divergence measures how far one probability distribution sits from another. Ambiguity shows up as volatility, and expected free energy becomes the mathematics of managing risk under uncertainty. He also explains why curiosity is mathematically required.
Why current AI cannot orient
We cover perceptual control theory, ecological psychology, and the difference between predictive and prospective control. We talk about flow states, the REBUS model and psychedelic-assisted therapy, autism and consciousness, and the Platonic space. REBUS is the model Karl developed with Robin Carhart-Harris for how psychedelics relax the brain’s prior beliefs.
The conversation closes on large language models. Current models do not encode uncertainty and cannot plan over counterfactual futures, so they cannot orient and cannot truly act. As I see it, the path forward is biomimetic, and Boyd’s sketch already points toward it.
Where is your own model of the world most uncertain right now?
Listen and watch:
YouTube:
Apple Podcasts:
Spotify:
No Way Out on YouTube:
https://youtube.com/@nowayoutpodcast
Written by Ponch Rivera with AI assistance in production. The full conversation is in the player below.