"Effective Computer Agents for Interacting With People"

Speaker: Kobi Gal
Date: Sunday, 21 December 2008
Time: 11am
Place: Ross 224

Abstract:

The range and complexity of human decision-making raises significant
challenges to the design of successful computer agents for interacting
with people. I'll describe work that addresses these challenges by
combining insights from behavioral economics and psychology.  I will
present a language for representing the different mental models people
may use to make their decisions, described as nodes in a graphical
network.  The language makes a distinction between agents' optimal
strategies and the way they actually behave in reality. I will show
how this language facilitates the design of computer agents that
represent and learn the social factors that affect people's
negotiation behavior.  In empirical investigations, agents using this
language to negotiate with other people were able to outperform
traditional game-theoretic equilibria.  I'll also talk about the need
for novel empirical frameworks in AI for investigating decision-making
in these mixed human-computer settings.