12.05.14 - 15.05.14, Seminar 14202

JA4AI – Judgment Aggregation for Artificial Intelligence

Diese Seminarbeschreibung wurde vor dem Seminar auf unseren Webseiten veröffentlicht und bei der Einladung zum Seminar verwendet.

Motivation

Judgment aggregation (JA) is the study of methods for amalgamating the views of several agents regarding the truth of a set of interrelated propositions. Originating in the humanities and the social sciences, particularly philosophy and economics, in recent years JA has started to also attract interest from computer scientists. The aim of this Dagstuhl Seminar is to give researchers across the contributing disciplines an integrated perspective on current research and emerging trends in JA, and to facilitate lasting collaboration across disciplinary boundaries.

JA was established as an independent research area in the early 2000's. In other areas of social choice, such as voting, preference aggregation, or fair division, research started with the proposal of concrete practical methods, and only later these methods have been analyzed using sophisticated theoretical frameworks, such as the axiomatic method. In JA research evolved in the opposite direction. The first half of the decade was marked by axiomatic studies, dominated by impossibility results and comparisons with the theory of preference aggregation. This work was largely conducted by researchers in the humanities and the social sciences. Due, amongst other things, to the potential for applications of JA in multi-agent systems and the parallels between concerns in JA and questions studied for some time in knowledge representation and reasoning, the second half of the decade witnessed an increase of interest in JA from researchers in computer science, and specifically in artificial intelligence (AI).

In AI, research on JA has since diversified significantly, with scholars pursuing a wide variety of different lines of research: JA and logic, JA and complexity theory, JA and belief merging, JA and argumentation theory, to mention a few. However, the awareness of these new directions amongst JA researchers outside of computer science is still very limited. At the same time, work in JA has continued in the humanities and the social sciences, also along many different lines of research: e.g., JA and deliberation, JA and strategic voting, JA and probabilistic opinion pooling. Many of those most recent results have not yet reached the AI community.

JA researchers from computer science and those from other fields each publish within their own discipline with virtually no cross-discipline cooperation on concrete projects. We believe that this is the right moment to pull together all these lines of research in JA, so as to enable cross-disciplinary cooperation and to chart the main research challenges for the coming years.

The seminar will feature both classical conference-style talks and a "rump session", a session of very short talks in which participants can quickly introduce a new research topic, a new idea, or a specific open problem. Most importantly, the seminar will provide plenty of space for discussion and collaboration.