| Περίληψη: | This doctoral dissertation studies contemporary issues of relevance and knowledge dynamics for intelligent agents. The research area that models knowledge in flux is that of Belief Revision (or Belief Change). In Belief Revision, as in many areas of Artificial Intelligence, relevance constitutes a crucial notion, both conceptually and computationally. The contributions of the dissertation to the aforementioned subjects are in the form of high-quality formal results and practical implementations oriented towards real-world problems.
Belief Revision is a young field of research that lies in the broader context of Knowledge Representation and Reasoning. The milestone of Belief Revision is a general and versatile formal framework introduced by Alchourron, Gardenfors and Makinson, known as the AGM paradigm (after the initials of its founders), which is, to this date, the dominant within the field. The AGM paradigm captures both axiomatically and constructively the process of rational belief revision.
Two main shortcomings of the AGM paradigm, as originally proposed, are its lack of any guidelines for relevance-sensitive and iterated belief revision. A major part of this dissertation is devoted to the establishment of important results concerning these two central sub-areas of Belief Revision. In particular, we study Parikh's relevance-sensitive axiom, as well as the most influential work addressing the problem of iterated belief revision, i.e., Darwiche and Pearl's approach.
In a more applied direction, several concrete "off the self" revision functions (operators), implementing the process of belief revision, have been proposed. In this work, we study a new and important proper subclass of AGM revision functions, called Parametrized Difference revision operators (or PD operators), which are natural generalizations of the well-known and intuitive Dalal's revision operator. PD operators bring us a step closer to the development of a successful AGM belief-revision system for real-world applications, due to their favourable properties, such as the low representational and computational cost, and the high expressivity.
Lastly, in this dissertation, a preliminary knowledge-based system capable of (formally) representing and reasoning about legal knowledge is developed. The work is implemented by means of the powerful and contemporary Answer Set Programming framework. The system has a plethora of extensions, and a variety of nice features (such as sufficient and reliable representation of the legal knowledge, handling non-monotonicity and exceptions, appropriate expressive power) that, in essence, reflect the nature of the Law. The overall approach constitutes an enhancing of the bond between Artificial Intelligence and Legal Science.
|
|---|
