The way in which the relationships between beliefs, goals, and intentions are captured by a formalism can have a significant impact on the design of a rational agent. In particular, what Rao and Georgeff underline about the relationships between goals and beliefs is that it is reasonable to require a rational agent not to allow goal-belief inconsistency, while goal-belief incompleteness can be allowed. We study a theoretical framework, grounded in possibility theory, which (i) accounts for the aspects involved in representing and changing beliefs and goals, and (ii) obeys Rao and Georgeff's requirement. We propose a formalization of a possibilistic extension of Bratman's asymmetry thesis to hold between goals and beliefs. Finally, we show that our formalism avoids the side-effect and the transference problems.
展开▼