Advanced Search   
  Look in
       Title    Full-text
  Date Range
RI Seminar: David Pynadath
Modeling Social Reasoning through Recursive, Decision-Theoretic Planning

David Pynadath

September 14, 2012, 3:30 PM, NSH 1305

Advances in artificial intelligence have led to subsequent advances in computational models of human decision-making and behavior. These new computational models provide powerful tools for social scientists, as well as more direct applications in social simulation, serious games, human-computer interaction, etc. Unfortunately, the richness and variety of human behavior often complicate and clutter such models, making it harder for designers to make use of them to create such applications. In this talk, I describe PsychSim, a general-purpose modeling architecture that uses theory of mind to capture the reasoning that people do when interacting with others, whether real or computer-generated. PsychSim's combination of decision theory with recursive models of others has supported the modeling of a wide variety of social phenomena, including influence theory, social norms, and self-deception. More importantly, PsychSim's unique algorithms facilitate authoring to the degree that non-computer scientists have successfully used it to model complex social scenarios for training games. By placing computational tools like PsychSim into the hands of the experts (social scientists, game designers, etc.), we can accelerate their use and their advancement along the most relevant directions of need.

Additional Information

Host: Paul Scerri

Appointments: Stephanie Matvey

Speaker Biography

David Pynadath is a Research Scientist at the Institute for Creative Technologies at the University of Southern California. His research has focused on developing an artificial intelligence framework that can model, simulate, and analyze social interaction, with special emphasis on the methods by which agents, both human and software, form and update beliefs about each other. In the context of software systems, he has analyzed the theoretical properties of optimal coordination and implemented an architecture capable of performing domain-independent teamwork among people, intelligent agents, and legacy software systems. In the context of human systems, he has built upon these algorithms to develop an agent-based simulation architecture, called PsychSim, that combines probabilistic reasoning and theory of mind to capture complex social phenomena. PsychSim has enabled the deployment of multiple simulation applications such as video games to teach people how to operate in social environments.