Abstract

Abstract: Sequential decision-making and planning problems are ubiquitous in engineering, economics, finance, artificial intelligence and robotics. Many of the hardest and most important problems involve long-term consequences of immediate choices, and therefore vast search-spaces. This talk proposes a new paradigm for general-purpose planning and decision-making that is appropriate for these large and challenging problems. The inspiration for this approach comes from the game of Go. This task is widely viewed as a grand challenge for artificial intelligence, which has thwarted traditional approaches to knowledge representation, learning and planning. Recently, a radically different approach has been developed, resulting in the first program, MoGo, to perform at human master level. This revolution in computer Go is based on three key principles: a) knowledge is represented by predictions about future experience, b) predictions are learnt directly from experience, c) planning is achieved by simulating experience. This talk proposes a more general application of this experience-based strategy. Just like Go, many real-world planning and decision-making problems have enormous search-spaces that are intractable to traditional search algorithms. Furthermore, also just like Go, in the majority of these problems, expert knowledge is unavailable or unreliable.