Self-Interested Agents and Abstract Argumentation

Speaker: Kate Larson

Since its introduction in the mid-nineties, Dung's theory of abstract argumentation has been influential in AI as it serves as a convenient model for reasoning about general characteristics of argument. Arguments are viewed as abstract entities with a binary defeat relationship between them. This has enabled extensive analysis of different argument acceptance criteria (semantics). However, most of this work ignores the fact that argumentation is often an adversarial process involving self-interested agents who may have conflicting preferences over the final status of arguments.

Recently, we have initiated the study of preferences and strategic behaviour in abstract argumentation. First, using concepts from social choice, we are able to discuss and compare the different abstract argumentation semantics in terms of the social desirability of their outcomes. Second, by introducing argumentation mechanism design, we are able to present formal results and characterizations of strategic behaviour in argumentation.

In this talk, I will provide a (gentle) introduction to abstract argumentation. I will then cover some of our results pertaining to the welfare properties of argumentation-based semantics. Finally, I will explain how we are using ideas from mechanism design in order to understand strategic behaviour in argumentation. I will conclude with pointing out several open questions.

This is joint work with Iyad Rahwan.

Material from this talk appeared in AAMAS 2008, AAAI 2008, and COMSOC 2008.