THE CHALLENGES OF REAL-TIME AL

The research agendas of artificial intelligence and real-time systems are converging as AI methods move toward domains that require real-tim e responses, and real-time systems move toward complex applications th at require intelligent behavior. They meet at the crossroads in an exc iting new subfield commonly called ''real-time AI.'' This subfield is still being defined, and the precise goals for various real-time AI sy stems are in flux. Traditionally, AI systems have been developed witho ut much attention to the resource limitations that motivate real-time systems researchers. However, as these AI systems move from the resear ch labs into real-world applications, they also become subject to the time constraints of the environments in which they operate. Rigorous d esign techniques developed by real-time systems researchers must be us ed to guarantee that a system will meet domain deadlines, even in wors t-case scenarios, particularly for mission-critical assignments. The a uthors would like to combine guaranteed performance methods with AI pl anning, problem-solving, and adaptation mechanisms to build a flexible , intelligent control system that can dynamically plan its own behavio rs. They describe an organizing conceptual structure, identify researc h goals, and specify some necessary steps for reaching them. They illu strate possible approaches with important applications that require th e best of both fields. A series of examples are provided from an inten sive care domain where an intelligent real-time control system could p rovide constant monitoring of patient needs.