Mechanism design for software agents with complete information

We investigate the mechanism design problem when the agents and the mechanism have computational restrictions. In particular, we examine how results in the mechanism design literature are affected when the social choice rule requires the mechanism to solve a computationally difficult optimization problem. Both dominant strategy and Nash implementation are considered for a multiagent version of the maximum satisfiability problem. We show that the best a mechanism can guarantee is that at least half of the maximum number of simultaneously satisfiable agents will be satisfied by the outcome. Our analysis highlights some of the difficulties that arise in applying results from mechanism design to computational problems. In particular, our results show that using approximation in multiagent settings can be much less successful than in traditional computational settings because of the game theoretic guarantees required of the outcomes.