Reinforcement Learning = Physics? The Mind-Bending Connection Between AI and Theoretical Physics

Explore the fascinating intersection of theoretical physics and artificial intelligence in this 40-minute talk that delves into the deep connections between Reinforcement Learning (RL) and Statistical mechanics. Discover how fundamental concepts like the Bellman equation, entropy-regularized expected reward, Shannon Entropy, and Boltzmann distribution create surprising parallels between physics and AI development. The presentation references important frameworks including soft actor-critique models, action-value functionals, state-value functions, and draws on Feynman's work to illustrate these connections. Beginning with insights from Northwestern University's paper "A Survey on Explainable Deep Reinforcement Learning," this mind-bending explanation challenges viewers to reconsider whether the future trajectory of AI might already be defined by the laws of physics.