A survey of real-time AI systems

As we move to the point where established techniques are incapable of squeezing out the last few percent of the potential efficiency that needs to be achieved, process automation has to look towards a new range of control philosophies. Plants are necessarily becoming more and more complex, and that the handling of this complexity presents major challenges. Expert systems are among the few tools coming out of the AI world which offer potential for incorporation into control systems and for supporting the overall structure of supervisory control systems. The technology is at an early stage, and we are really only just beginning to understand its consequences. We must not be intimidated by the technology, but should simply regard it as just another way of programming our computers! This paper addresses the question of using the technology in hard real-time situations. To provide a background to this, an introduction is given to the `real´ requirements of real-time systems. These requirements are matched to expert systems in particular, and some pointers are given as to how current systems are being modified. However, radically new architectures are, of necessity, having to be developed