07. – 12. August 2005, Dagstuhl Seminar 05321
Belief Change in Rational Agents: Perspectives from Artificial Intelligence, Philosophy, and Economics
Auskunft zu diesem Dagstuhl Seminar erteilt
The area of belief change studies how a rational agent may maintain its beliefs when obtaining or perceiving new information about the environment. This new information could include properties of the actual world, occurrences of events, and, in the case of multiple agents, actions performed by other agents, as well as the beliefs and preferences of other agents. Not surprisingly, this area has been of interest to researchers in different communities.
The initial research in belief change came from the philosophical community, wherein belief change was studied generally from a normative point of view (that is, providing axiomatic foundations about how rational agents should behave with respect to the information flux). Subsequently, computer scientists , especially in the artificial intelligence (AI) and the database (DB) communities, have been building on these results. Belief change, as studied by computer scientists, not only pays attention to behavioural properties characterising evolving databases or knowledge bases, but must also address computational issues such as how to represent beliefs states in a concise way and how to efficiently compute the revision of a belief state. More recently, the economics and game theory community, in particular the emerging field of cognitive economics , has become active in belief change research, adopting a normative point of view, like philosophers, but paying more attention to the "cognitive plausibility" or "fitness" of the belief change operators.
The goal of the seminar is to bring together researchers from these areas. This would allow the identification and addressing of problems of common interest in this highly challenging and relevant area, as well as an exploration of ways in which one area may contribute to another.
Goals and Content of the Seminar
The area of belief change can be regarded as originating in the philosophical logic community. This work provided abstract, formal, and precise specifications of desirable properties for belief change operators, as well as the identification of distinct types of change. However, this research says nothing about specific implementable operators nor computational issues -- issues of fundamental importance to computer scientists. Researchers in artificial intelligence and computer science have followed up on these latter issues, as well as developed other specific operators (addressing e.g. sensor fusion and belief base merging) and examined their complexity characteristics. In artificial intelligence, the relatively recent emergence of the field of cognitive robotics , which is concerned with endowing artificial agents with cognitive functions that involve reasoning, for example, about goals, actions, the states of other agents, collaboration and negotiation, etc., has given impetus to the development of computational operators for belief change and the identification of issues arising from concrete, evolving sets of knowledge. As well, more recently, economists have been using work in belief revision, and applying it to notions of mistaken and changing beliefs among interacting and negotiating agents. Such work is also of obvious interest to researchers in artificial intelligence.
To date, there has been limited interaction among these communities. Clearly however there are deep problems of common interest, and results in one area will contribute to another. We have already mentioned that research in economics has made use of the work from the philosophical community, and that such results will be of use to researchers in AI. As well, contributions may also flow back from economics to research in the foundations of belief revision: For example, recently it has been suggested that that economic principles (dealing with choice, preferences, and utility) may provide a more appropriate foundation for belief change. Computational issues raised and addressed by researchers in computer science and AI will be of use to economists addressing related problems; as well such work can contribute to the other areas by further elucidating the abstract area of belief change, as well as providing implementations and identifying philosophically-interesting "pragmatic" or "practical" problems.
Thus we see researchers in three broad areas (philosophy and logic, artificial intelligence and computer science, and economics and game theory) addressing highly related (in some cases, the same) problems, in which work in one area in all likelihood will benefit research in another. Hence for the Dagstuhl seminar, we feel that there would be valuable interactions and contributions that would be anticipated by bringing people together in these areas.
We found the workshop successful, especially on the following two achievements: first, the seminar made participants aware of a commonality of interests across different disciplines; second, it suggested new directions for research that will probably be taken up by researchers in the next couple of years.
Where is the field going? We can mention at least two emerging issues:
- the field is broadening both with respect to theoretical underpinnings, and so begining in incorporate notions from game theory and social choice theory. As well, it is broadening wrt application areas, moving beyond traditional areas in AI and database systems, to including areas in description logics, the semantic web, and in economics.
- As well, there is an emerging focus on epistemic notions having to do with communicating, negotiating, competing, and collaborating agents in belief change. Dynamic epistemic logic seems to have an important role to play here.
Moreover, it looks like belief merging and iterated belief revision are still hot topics and will remain so for the next few years.