Explicit Models in Intelligent Interface Design

Abstract

AI has made significant advances in the design of graphical knowledge-based interfaces. One type of knowledge-based interface, the graphical model editor, extends the "toolkit notion of interface", exemplified by Bill Budge's Pinball Constructor Set 1. In these interfaces, "the desired operations are done simply by moving the appropriate icons onto the screen and connecting them together. Connecting the icons is the equivalent of writing a program... There are no hidden operations, no syntax of command names to learn." 2) The end user can quickly, and with minimal training, build up a description of a complex structural model, run that model to observe its behavior, and then modify the structure if desired. The user of this class of interface never needs to be aware of the complexities of the internal representations that actually support the model.AI has contributed flexible tools for building these interfaces such as Impulse 3) the steamer graphics editor, 4) and KEE pictures TM5, as well as representation techniques that facilitate mapping a complex domain model onto a graphical interface. However, the class of text editors known as structure editors can also serve as a useful analogy for extending our understanding of graphical model editors.In this paper, we will first define what we mean by graphical knowledge-based structure editors and then discuss one class of structure editors, those that can be built using SimKit TM, a general tool for building model editors. We will then speculate about the specific types of knowledge typically represented in knowledge-based structure editors and the appropriate graphical metaphors for presenting that knowledge.