PHI : a logic-based tool for intelligent help systems

Abstract

We introduce a system which improves the performance of intelligent help systems by supplying them with plan generation and plan recognition components. Both components work in close mutual cooperation. We demonstrate two modes of cross-talk between them, one where plan recognition is done on the basis of abstract plans provided by the planner and the other where optimal plans are generated based on recognition results. The examples which are presented are taken from an operating system domain, namely from the UNIX mail domain. Our system is completely logic-based. Relying on a common logical framework--the interval-based modal temporal logic LLP which we have developed--both components are implemented as special purpose inference procedures. Plan generation from first and second principles is provided and carried out deductively, whereas plan recognition follows a new abductive approach for modal logics. The plan recognizer is additionally supplied with a probabilistic reasoner as a means to adjust the help provided for user-specific characteristics.