Colloquium: A Computational Framework for Interactive Decision Making with Applications in Strategic Tasks

6 June 2018

Abstract. Sequential tasks such as Rosenthal's Centipede Game and Berg's Trust Game have long offered insights into human strategic thinking and behavioral economics. In particular, cooperation and trust, while precluded in these games by Nash equilibrium, is often observed emerging among the players. Simultaneously, the players exhibit low levels of recursive reasoning (reasoning of the type "I think that you think that I think.") in general contexts. Recently, cognitive and neuro-scientists have started to computationally model this complex behavior using generative process models. In this talk, I will selectively review some studies involving the Centipede and Trust Games and discuss the observed behavior. Next, I will focus on how the behavioral data from these studies was computationally modeled. To properly understand the modeling, I will explain the framework for normative decision making, called the Interactive POMDP, that was utilized in the modeling. I will discuss the ways by which the framework is adapted to model descriptive reasoning and acting. The resulting model is both generative and parameterized, and offers a process account of the observed behavior with good accuracy. I will conclude my talk by briefly outlining another application of this framework in human-robot interaction.

Monday, 06 June 2018, 15:00, Seminar Room, Building W34, Institute for Systems Neuroscience, UKE Campus

Speaker: Prof. Dr. Prashant Doshi, AI Institute, University of Georgia.