Cooperating with Machines Rogers 226 Fri, 04/15/2016 - 10:00am
Jacob Crandall Associate Professor, Department of Electrical Engineering and Computer Science, Masdar Institute of Science and Technology Abstract: Since Alan Turing envisioned Artificial Intelligence, a major driving force behind technical progress has been competition with human cognition. Historical milestones have been frequently associated with computers matching or outperforming humans in difficult cognitive tasks (e.g. face recognition, personality classification, driving cars, or playing video games), or defeating humans in strategic zero-sum encounters (e.g. Chess, Checkers, Poker, or Jeopardy!). In contrast, less attention has been given to developing autonomous robots that establish mutually cooperative relationships with humans even when the preferences of humans and robots are, to some degree, in conflict. A main challenge has been that human cooperation does not appear to require sheer computational power, but rather relies on intuition, cultural norms, emotions and signals, and pre-evolved dispositions toward cooperation, common-sense mechanisms that are difficult to encode in robots. In this talk, I will describe my research group’s efforts to develop algorithms that can establish and maintain cooperative relationships with people. In particular, I will describe a new learning system that combines a state-of-the-art machine-learning algorithm with novel mechanisms for generating and acting on signals. Given only a description of the game, the resulting learning system cooperates with people and other robots at levels that rival human cooperation in two-player repeated interactions. It does so without pre-programming of well-known, game-specific strategies, thus enabling human-AI cooperation in scenarios previously unanticipated by algorithm designers. Bio: URL: http://eecs.oregonstate.edu/colloquium/cooperating-machines _______________________________________________ Colloquium mailing list [email protected] https://secure.engr.oregonstate.edu/mailman/listinfo/colloquium
