Qualitative Process Theory: Twelve Years After

The scientific goal of artificial intelligence is to understand minds by trying to build them. To make progress towards this goal requires careful decomposition . An especially productive strategy has been to focus on the knowledge and reasoning required for a particular class of domains or tasks. Some domains or tasks have special features which make them tractable, while still providing generalizable insights . If the domains and tasks have practical value then so much the better, for then scientific progress and economic benefits can go hand-in-hand . Qualitative physics is such an area. Reasoning about the physical world is clearly central to intelligences (human or machine) . Moreover, it encompasses a variety of tasks and skills, thus providing a range of interesting problems. Physics and mathematics provide us with clues about what special constraints might help make such reasoning tractable. And the potential economic impact of intelligent computer-aided engineering systems is clearly enormous . Indeed, qualitative physics is one of the most exciting areas in AI today. An important event in the growth of qualitative physics was the 1984 special issue of Artificial Intelligence edited by Bobrow . My paper in that issue, "Qualitative process theory", was one-half of my Ph .D . thesis; the other half described algorithms for implementing QP theory and results obtained with them . This essay summarizes the context of that work and the main contributions of QP theory . A cautionary point about basic research