Decision Under Insufficient Evidence: A Scalable Probabilistic Way

Some problematic cases, such as collective defeat and odd-length defeat cycles, which tend to be handled incorrectly by all of the current theories of no monotonic reasoning, including default logic and circumscription, have been well recognized in the literature. Although a powerful argument-based approach in the automated defeasible reasoner OSCAR has been proposed and they claim that this theory is able to reason correctly for the problems above all, but we don't consider it to be true completely through careful investigation. It seems to be the consequences of disconnection between epistemic reasoning and practical reasoning and not considering the possible consequences of the decision and individual preferences sufficiently. Following these observations, we propose a scalable probabilistic approach based on Bayesian decision theory that can solve all of the above paradoxes properly and has successfully been used in web of trust and knowledge integration for semantic Grid.