Search for an ideal reasoning system: speci cation of the minimal requirements

In this paper we will present criteria for critical evaluation of plausible reasoning systems. We have taken the rational reasoning as our point of comparision: an ideal reasoning system should be able to represent all human knowledge and perform all reasoning tasks, which can be rationally performed. The four main criteria are exible ways of representing uncertainty, the ability for both deductive and inductive reasoning, nonmonotonic reasoning and the coherence of the system. After determining the criteria we will evaluate three existing reasoning systems according to our criteria: the Bayesian networks, the Dempster-Shafer theory and the uncertain extensions of the Assumption-based Truth Maintenance System. In addition, we will brie y report practical experiments, which demonstrate some of the restrictions and strengths of the existing reasoning systems.