Communicating With High-Level Plans

This paper provides support for the following points regarding the design of intelligent interfaces (or more generally, intelligent dialog systems): 1) Constraints of human dialog processing, including limited attention span and partial understanding, should be dealt with explicitly in design of intelligent dialog systems; 2) An intelligent dialog system should be a partly autonomous agent, though accountable to its users; 3) An intelligent dialog system should be designed to incorporate explicit active models of tasks, users, the system itself and the process of dialog with the user; 4) Models should be built and maintained both for types of tasks, users and dialog processes and for specific instances for tasks, users and dialogs. In particular, an intelligent dialog system should model users in terms of their individual human characteristics; 5) The configuration of models maintained by an intelligent interface should be self referencing. That is, the system's model of itself, which is a component of the model configuration, should be able to refer to the model configuration as a whole, and should be able to access the assertions in that configuration; and 6) Dialog processes should be modeled from a first-party (system) view and the system's second-party view of the user, rather than from a third-party (observer's) view.