Real-Time AI Systems: A Definition and An Architecture

Speed alone is insufficient for real-time performance. We define real-time performance in terms of speed, responsiveness, timeliness, and graceful adaptation. We claim that all four aspects are essential if a system is to support real-time problem-solving. We also present a distributed knowledge processing architecture based on the lackboard paradigm that addresses all aspects of real-time performance. Primary attention was given to flexibility of behavior without compromising on the efficiency of implementation so that the applicability of the architecture to an application may be experimented with. Performance metrics are crucial for validating real-time performance, and form an integral component of a real-time system. In this paper, we present performance metrics for responsiveness and timeliness at the architecture level.