In this episode, hosts unpack computational rationality by contrasting model-free and model-based learning through vivid examples — from a crashing delivery drone to Thorndike’s cats and Tolman’s cognitive maps. They explain how habits form as amortized inference, why planning saves costly trial-and-error, and how active inference reframes goals as prediction-driven behavior.

The conversation covers reinforcement learning basics, the tradeoffs between fast habits and flexible planning, Friston’s free energy perspective, and the role of bounded optimality and policy complexity in shaping human decision-making. Practical takeaways include a "policy audit" for personal habits and a coding exercise to explore model-based vs. model-free agents.