1988 Reports

Finding a Better Way: Choosing and Explaining Alternative Plans

Wolz, Ursula

Many users of interactive computing environments have been confronted with the dilemma: "I know there must be a better way to do what I'm doing, but 1 don't know what it is." This paper describes a solution to this problem and several related ones through an automated consultant called GENIE, a system that answers users' questions about how to accomplish tasks. Our approach is based on the principle that the best plan to tell a user is not always computationally optimal, but is the plan most suitable to what the user already knows in the current context. To find a best plan, GENIE explores a declarative representation of goals that explicitly encodes alternative plans for goals, and the semantic relationships between alternatives. Both an Expert and a User model are represented in this manner, which allows us to abandon stereotypes of user expertise and functional difficulty of the domain constructs. The criteria for each choice of sub-step of a plan is based on two sets of heuristics applied to the semantic relationships encoded between plans. The first set is dependent upon what the user is currently doing, the second on what the user has done in the past.