Chapter 11 - Temporal Reasoning in Artificial Intelligence

Abstract

In one way or another, every area of artificial intelligence (AI) has to do with time. Medical diagnosis systems reason about the time at which a virus infected the blood system. Device troubleshooting systems look at how long it takes a capacitor to saturate. In automatic programming, the time at which a variable becomes bound is important. In robot planning, one wants to achieve one goal before another to meet deadlines, and so on. In qualitative physics, the concept of time is essential as well. One can identify several classes of tasks in AI that require reasoning about time: (1) prediction, (2) explanation, (3) planning, and (4) learning new rules. These classes of tasks, though related, have given rise to by and large disjoint fields of research. These disjoint research areas can be unified to some extent by providing a uniform framework for temporal reasoning. The somewhat mythical area of temporal reasoning aims to provide such a framework. The representation of temporal information and reasoning about such information requires a language that can capture the concept of change over time and can express the truth or falsity of statements at different times. This language should not only be well-defined but also have a clear meaning. This has led researchers to develop temporal logic. The passage of time is important only because changes are possible with time. This chapter explains two different approaches to reasoning about change: change-based and time-based. The chapter presents an introduction to a representative temporal logic with formal syntax and semantics. It also provides an overview of the problems and the advances made in nonmonotonic temporal reasoning.