ICES: Small: Mechanism Design for Highly Anonymous Environments

Duke University, Durham NC

Investigators

Abstract

In mechanism design, the goal is to design rules that lead to desirable outcomes even when agents behave strategically. This project focuses on a specific type of strategic behavior: a single agent can pretend to be multiple agents, and thereby participate in the mechanism more than once. One approach to address this is to design the mechanism so that this behavior is always suboptimal for the agent. Other approaches include: making it costly to obtain additional identifiers; doing some limited verification of identities; detecting false identifiers based on the social-network structure among the agents; and designing methods to sample from the agents in an unbiased way. These approaches require a blend of techniques from computer science and economics. Internet-based mechanisms such as online rating systems and online elections, as well as mechanisms for electronic commerce, play an increasingly important role in the economy and social infrastructure. It is often easy to participate multiple times in such mechanisms, leading to distorted results and economic inefficiencies. Hence, the research under this project has the potential to make Internet-based mechanisms more meaningful and efficient.