System initiative influenced by underlying representations in mixed-initiative planning systems

Mixed-initiative artificial intelligence systems are ones where the system can take the initiative to solicit further input from the user or where the user can take the initiative to interrupt the processing of the system to provide additional input (see (Haller, McRoy, ~ Kobsa 1999), (Cox 1999)). Mixed-initiative systems contrast with other AI systems which perform automated reasoning given input from a user, producing output without further interactivity. To date, the design of mixed-initiative AI systems has been largely done independently by researchers in various subareas of AI, including intelligent tutoring systems ((Aist 1997), (Lester, Stone, & Stelling 1999)), planning systems ((Allen 1994), (Cox & 1997), (Surstein & McDermott 1996)) and interface agents ((fleming & Cohen 1999), (Rich & Sidner 1998), (Cesta & D’Aloisi 1999)). The aim of our search is to develop some general guidelines to direct the design of mixed-initiative AI systems in a more principled manner. To this end, our plan is to first study the requirements of various kinds of mixed-initiative applications and to then identify some guiding influences for successful system design. In particular, we hope to provide some concrete answers about what characteristics of an application domain make it particularly well-suited to a mixedinitiative approach, and what factors must be considered by a system in determining whether or not it is appropriate to take the initiative in a given situation. We also intend to build on the limited existing work on evaluating the effectiveness of mixed-initiative AI systems. A possible first step for our general framework will be the development of a formula for a system to use in determining whether or not to take the initiative in a given situation, to ask the user for assistance. Only preliminary thought has been given to this so far, but several factors have been identified which should be incorporated into any such formula: