Evolutionary mechanism design with strategic agents

Mechanism design is the inverse problem of game theory, it specifies the consequences and searches for the rules of the game leading to those consequences. Mechanisms have traditionally been designed manually for classes of problems. Conitzer and Sandholm introduced the automated mechanism design (AMD) approach where the mechanism is computationally created for the specific problem instance at hand. AMD solves mechanism design as optimization problems and uses linear programming to solve them. Since the complexity of the linear optimization problem is exponential in the number of agents, this method is tractable only for small numbers of agents. In this paper we propose a method for AMD by using fast evolutionary programming with swarm directions. It yielded several mechanisms for a divorce settlement scenario and public goods problems better than the public literature.