Distributed consensus mechanisms for self-interested heterogeneous agents

In a previous paper [8] we introduced the Clarke Tax [2, 3] as an appealing way to achieve a consensus in a multi-agent environment. As noted there, however, the Clarke Tax has several drawbacks from the viewpoint of DAI, such as the fact that participating agents can maintain no privacy regarding their preferences, and the need for a central vote-counting agent. In [7] we considered the first of these problems, namely how agents can maintain a certain amount of secrecy regarding their preferences, and still converge to a socially optimal choice. In this paper, we address another of these Clarke Tax drawbacks, the need for a central vote-counting agent. We present several distributed voting schemes that do away with the need for a central vote counter. This improves the reliability of the system, and eliminates a potential system bottleneck. By distributing the vote counting, we also help maintain agent privacy-fewer agents need to know how an individual voted. We also consider how the Clarke Tax could be modified to give different agents different amounts of influence over the group decision.