Artificial intelligence affects how we understand the behavior of machine learning systems. Stefano Soatto, VP of Applied Science, Amazon Web Services, explains how ideas from information geometry shape emerging theories of how these artifacts work. Soatto examines the natural gradient, the connections between geometry and concepts such as probability distributions, entropy, mutual information, and KL divergence, and the challenge of defining information in trained models, helping clarify how reasoning and learning can be understood in the era of AI. Series: "Kyoto Prize Symposium" [Science] [Show ID: 41494]