AIAI Seminar - 16 May 2022 - Talk by Levi Lelis

Speaker: Levi Lelis

Title: Learning from Machines

Abstract:

We recently witnessed tremendous achievements of machine learning systems. For example, AlphaZero achieved superhuman performance on challenging games such as Go, Chess, and Shogi. Yoshiharu Habu, a 9-dan Shogi player stated the following about AlphaZero: “Some of its moves go against Shogi theory and seem to put AlphaZero in a perilous position. But incredibly it remains in control of the board. Its unique playing style shows us that there are new possibilities for the game”. While systems such as AlphaZero are able to learn strong and novel strategies for difficult problems, the generated solutions are encoded in opaque models, which are often hard for us to understand. Lack of interpretability hinders our ability to learn from these systems. In this talk I will discuss different methods that allow people to learn from intelligent systems. I will present a study case of people learning how to strategize in the Olympic sport of curling from justifications of the decisions an intelligent system makes. I will also present a study case where people learn how to solve puzzles from machine-generated curricula.

Bio:

Levi is a Fellow and Canada CIFAR AI Chair at Amii, an Assistant Professor in the Department of Computing Science at the University of Alberta, and a Professor on leave from the Universidade Federal de Viçosa, Brazil. He received his Ph.D. in Computing Science in 2013 from the University of Alberta. Levi has co-authored more than 45 refereed papers at prestigious artificial intelligence venues.