The reliability of reasoning with unreliable arguments

Shafer’s theory of evidence has found much interest as a means to represent and manage uncertainty in reasoning. It is shown that an important part of this theory can be recast as a theory of reasoning with unreliable arguments. This places the problem within the framework of the reliability theory of binary, monotone systems. Methods from this well developed theory can thus be used in the context of reasoning under uncertainty. This opens new computational perspectives for combining evidence. In particular, it allows for a ‘backward chaining’ approach in judging hypotheses in the light of unreliable information. As a consequence, mechanism to explain the judgments of hypotheses can be designed. The approach will be illustrated especially by the problem of reasoning in inference networks with uncertain rules and uncertain facts.