Mixed-Initiative Dialogue Systems for Collaborative Problem-Solving

This paper describes our motivation for and approach to the design and implementation of mixed-initiative dialogue systems for collaborative problem solving. Our model allows true mixed-initiative dialogue, where goals (and other joint commitments) can come from either the user or the system. The system’s behavior with respect to its commitments is driven by a formal model of collaboration based on a theory of joint intention which also allows the system to interpret the utterances and actions of the user in a uniform manner. Introduction and Motivation Our goal is the design and implementation of collaborative assistants that help people solve problems and get things done. We are primarily concerned with systems that interact using spoken natural language dialogue since (a) this is a very efficient means of communication; (b) it requires little or no training; and (c) it gives us the greatest insight into the nature of human communication and collaboration. Other modalities such as graphical user interfaces and visualizations such as graphs and tables are also possible, with some caveats that we elaborate towards the end of this paper. A good assistant is necessarily one that can not only respond to your initiatives but can also take initiative itself. We’ve all had the bad experience of working with someone who had to be told everything they needed to do. Similarly, a system that rigidly controls the dialogue, such as a telephone menu system, can hardly be considered to be working with you, much less for you. We like Eric Horvitz’ characterization of mixedinitiative systems: I shall use the phrase to refer broadly to methods that explicitly support an efficient, natural interleaving of contributions by users and automated services aimed at converging on solutions to problems.(Horvitz 2000) The standard motivations for mixed-initiative systems center on making the most of the different abilities of the user and the system. Who would argue with this? Copyright c © 2005, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. Typing