Refinement: a tool to deal with inconsistencies

In traditional Belief Revision approaches, new information is accepted unconditionally. In such models the agent either corrrects his knowledge by introducing a new conflicting information or simply rejects it. This paper proposes an alternative way where, in presence of new conflicting data, the agent refines his knowledge by accepting the new conflicting information under certain conditions. Such operation, called here refinement, provides an interesting tool to avoid contradictions in normative systems and is a first step towards a rigorous treatment of how courts provide fresh solutions to circumstances which were unpredicted by legal statutes. It is an alternative to defeasible conditionals approaches much in the spirit of Alchourrón's defense of revision as the proper account of ampliative resoning [1]. The paper introduces a constructive model for refinement based on the power of the AGM contruction [3] and inspired by general ideas of selective revision put forward by Fermé and Hansson [6]. A refinement operator @@@@ is defined and an axiomatic characterization is provided with a representation theorem.