Cooperative Hybrid Systems

Matthias Gutknecht, Rolf Pfeifer, Markus Stolze

Abstract: Recently there has been much criticism in the AI community that knowledge based systems are not situated. We argue that trying to provide for situatedness in a conventional system will lead to the so called model explosion cycle and that for most application environments adaptivity is needed for situated behavior. Cooperative systems are a solution to the model explosion cycle where adaptivity is delegated to the user. A hybrid symbolic/connectionist system offers self tuning capabilities (and therefore adaptivity) but can't cope with the model explosion cycle. Integrating both approaches into a cooperative hybrid system leads to a much more situated behavior than conventional systems can achieve. Our approach is illustrated using a real-life expert system in the domain of technical troubleshooting. Ongoing practical tests indicate that a cooperative hybrid design presents an attractive architecture for knowledge based systems.