|
Abstract : |
Abstract. Traditional decision theory has assumed that agents have complete, consistent and readily available beliefs and preferences. Obviously, even if an expert system has complete and consistent beliefs, it cannot have them readily available. Moreover, some beliefs about beliefs are not even approximately computable. It is shown that if all players have complete and consistent beliefs, they can compute approximate beliefs about beliefs of any order by considering events arbitrarily close in some well-defined sense to the ones in question. 1., |
