In recent years,AI researchers and system developers have grown increasingly interested in task-oriented approaches to problem solving. One task, temporal reasoning, is pervasive in many AI activities, including diagnosis, planning, scheduling, temporal database management, and natural-language understanding. These activities would benefit from a temporal knowledge server that could deal efficiently with various types of temporal information. Indeed, specialized temporal-information managers have emerged, and AI researchers have proposed several approaches for dealing with time in problem solving.

Later (layered temporal reasoner), our general-purpose manager of temporal information, fills such a need. It exhibits the following characteristics:

Users can exploit Later in problem solving by adopting a modular approach that loosely couples our system with other modules. After describing Later's architecture and operations, this article demonstrates its usefulness with a concrete example involving temporal model-based diagnosis.