Industry Connections: Meeting the Challenge of Cracking Open Large-Scale AI with David Bau

About the Talk: What can humanity learn when we crack open black-box AI systems and look inside? In this talk David will examine some recent insights from interpretable machine learning, asking how large-scale AI systems organize knowledge, Theory of Mind, and a "dual route model of induction" that uncovers a separation between token, concept, and functional representations during in-context learning.  After discussing recent work from the Bau lab as well as some inspiring results across the field, David will also discuss policy questions confronting machine learning: how will we preserve human agency in a world of "superhuman AI;" is interpretability essential for long-term progress in the field; and if so, what does it mean about the way we should be structuring our industry today? About the Speaker: David Bau is an assistant professor in the Khoury College of Computer Sciences at Northeastern University, based in Boston. Bau's research focuses on human-computer interaction and machine learning. Before joining Northeastern, he worked as a software engineer at Google, BEA, and Crossgain. He has been published in journals such as CVPR, NeurIPS, ICCV, ECCV, and SIGGRAPH. Outside of research, Bau enjoys astronomy and puzzle collecting.