Personal learning apprentices

Abstract

We believe that learning apprentice systems may make practical a number of personalized work assistant applications which are now impractical due to the prohibitive costs of knowledge base development and maintenance. CAP is a prototype learning apprentice for one such application: personal calendar management. While CAP is still a prototype system with only a fraction of the capabilities we envision, it is already in routine use by one secretary in our environment, collecting training data on a regular basis. Our preliminary learning experiments indicate that CAP's inductive learning methods are sufficient to automatically acquire rules comparable in performance to manually created rules, for the two features of meetings we have tested thus far. We plan to continue development of CAP, and to distribute copies to a number of users in our environment in order to test its ability to customize itself to a variety of individuals.