Ulric B. and Evelyn L. Bray Social Sciences Seminar

Abstract: How can we use machine learning to advance science? For many disciplines, answering this question is an active area of research e.g. physics, material sciences or biology? In this talk, I will focus on a category of sciences that are qualitatively different from the usual suspects, what I will call the inscrutable sciences. That category - which includes fields such as behavioral economics or medicine - raises new challenges. First, these fields often rely on "theories" that are informal, and sometimes even implicit. Second, even when formal and explicit, these fields have theories that are far from perfect. As a result, the role of algorithms can be quite distinct - and requires new computational tools and frameworks. In this talk, I will describe our efforts to state and solve these algorithmic problems, as well as applications to concrete problems.

Joint work with Jens Ludwig and Ashesh Rambachan.