Preference Reasoning and Aggregation: Between AI and Social Choice

Preferences are ubiquitous in everyday decision making. They should therefore be an essential ingredient in every reasoning tool. Preferences are often used in collective decision making, where each agent expresses its preferences over a set of possible decisions, and a chair aggregates such preferences to come out with the "winning" decision. Indeed, preference reasoning and multi-agent preference aggregations are areas of growing interest within artificial intelligence. Preferences have classically been the subject also of social choice studies, in particular those related to elections and voting theory. In this context, several voters express their preferences over the candidates and a voting rule is used to elect the winning candidate. Economists, political theorist, mathematicians, as well as philosophers, have made tremendous efforts to study this scenario and have obtained many theoretical results about the properties of the voting rules that one can use. Since, after all, this scenario is not so different from multi-agent decision making, it is not surprising that in recent years the area of multi-agent systems has been invaded by interesting papers trying to adapt social choice results to multi-agent setting. An adaptation is indeed necessary, since, besides the apparent similarity, there are many issues in multi-agent settings that do not occur in a social choice context: a large set of candidates with a combinatorial structure, several formalisms to model preferences compactly, preference orderings including indifference and incomparability, uncertainty, as well as computational concerns. The above considerations are the basis of a relatively new research area called computational social choice, which studies how social choice and AI can fruitfully cooperate to give innovative and improved solutions to aggregating preferences given by multiple agents. This talk will present this interdisciplinary area of research and will describe several recent results regarding some of the issues mentioned above, with special focus on robustness of multi-agent preference aggregation with respect to influences, manipulation, and bribery.