Reducing computation by unifying inference with user interface

User interfaces are becoming an integral component of usable interactive artificial intelligence systems. A high-bandwidth graphical interface between the user and the computer's representation of knowledge greatly facilitates intelligent human/computer interaction. When user input or inference modifies knowledge base assertions, computationally expensive cascaded updates often propagate through the underlying knowledge representation. To communicate these changes to the user, the graphical display must also be updated. The speed of knowledge base update is crucial in Al interfaces. Real-time global update is often infeasible due to the richness and complexity of knowledge bases. Interestingly, the user's display usually presents only a fraction of all the available internally represented information. To rapidly show the user the global effects of changes, knowledge base update can be restricted to just those computations required for maintaining an accurate display. We have developed an integrated approach to inference and user interface which offers a new programming paradigm for the construction of Al interfaces. The methodology: • greatly simplifies the design, construction and runtime execution of user interfaces to knowledge bases; • reduces inference computation by user interface-constrained lazy evaluation: only what the user can observe is computed; • integrates inference and user interface into one system specifically tailored to the Al user's immediate needs. This publication was supported in part by grant R29 LM 04707 from the National Library of Medicine. Table of