In this talk I will discuss the key connections and differences between the model-based andmodel-free paradigms from the perspectives of reinforcement learning and generative AI. I will arguethat establishing a sufficiently accurate model is both impossible and unnecessary for the ultimatepurpose of making optimal decisions, but there is some quantity, one that is an aggregate measure ofthe model parameters and control actions, that needs to be learned and can indeed be learned efficientlyin a data-driven way.
