Maria Vanina Martinez Posse

Ricardo O. Rodriguez

Since the 1980s, the AGM model of revision and contraction operators in belief change
theory has been adopted in artificial intelligence to address the problem of updating
knowledge bases with potentially inconsistent information. This model provides a formal
path to manipulate the update in the presence of inconsistencies in a precise and clear
semantic way, combined with a realistic perspective from a computational point of view.
Over time, the AGM model has been generalized to apply in various contexts, for example,
prioritized operators for dynamic environments, such as update and erase; prioritized
multiple change operators, such as package and choice; non-prioritized operators, such
as credibility-limited revision, shielded contraction, or filtered revision; and operators for
non-classical logics.
This thesis proposes a homogeneous approach to analyze this diversity under a unified
theoretical framework. We introduce a possible worlds semantic, independent of any
underlying logic, where beliefs are represented solely as a set of worlds, without relying
on a specific syntax. We then adapt several known models to this framework and propose
a family of non-prioritized operators encompassing the previous proposals. Finally, we
demonstrate that our theoretical framework effectively homogenizes and generalizes the
classical, finite, and multiple proposals known for classical propositional logic.