Adaptation to Unfamiliar Task Conditions: A Shared Problem Between Artificial Intelligence and Psychology

Stellan Ohlsson

University of Illinois at Chicago, USA

Fri Mar 05 15:10:00 NZDT 2004 in Room 031, MSCS

Abstract

In this presentation I will pose the very general problem of cognitive change, and briefly outline three approaches that we have pursued in my research group at UIC. I will then develop one of these in more detail: How can an intelligent system adapt to unfamiliar task conditions during the execution of task? A task environment generates positive and negative feedback, but how can an intelligent agent make use of that feedback to improve its performance? I will pay particular attention to the problem of how to learn from errors, i.e., from negative feedback. We have developed several computer simulations to explore different facets of this problem, and I will present and discuss a handful of simulation results. The presentation will end with some general reflections on adaptation through feedback.