סמינר מחלקתי

Human-Technology Exploration: how do curious agents interact with learning systems?

Dr. Goren Gordon - MIT

Abstract:

People interact with novel technology all the time. How do they explore it? How do new learning systems explore their users to better understand and react to them? In this talk I present recent results from the study of curiosity, both human and artificial. To better understand basic curiosity-driven behaviors, we first studied rodents' exploration of a new environment, where we showed that curious animals optimize their novelty signal-to-noise ratio, so that their novelty input is high, yet stable. Incorporating this insight into an information-theoretic model of hierarchical curiosity loops and novelty management, resulted in a minimal model that can explain and predict many observed behaviors. We showed that people exploring with artificial whiskers followed these same principles. The same model, implemented in robots, results in curious robots that learn about themselves and people interacting with them. Finally, results from a recent study on how children interacted with robots, where human and artificial curiosity are combined in a literacy educational setting, indicate that children can “catch” curiosity from a curious robot in a selective manner. Future work on the study of how adults curiously explore novel technologies, and how these systems explore and personalize to their users concludes the talk.