Dynamic expert systems

Introduces the notion of a dynamic expert system. Shows that by restarting a static expert system periodically it is possible to cope with dynamic environments. This quasi‐static approach to dynamics is suitable if the environment is changing slowly enough in comparison with the inference engine operation and the user reaction time. Implementation of this scheme in “pure” shell meets no difficulties. However, in practice some problems may occur due to the side‐effects in rules and attached procedures. These problems and their relation to classical AI issues are considered in detail. The system was applied to the task of seismology forecast, which contains dynamical factors of both a precise and a heuristic nature. The resulting dynamic expert system never stops, occasionally interrogating the user if it suspects that some of the previously entered data are obsolete. In this sense the computer system behaves as a “live creature”.