Organizing unreliable decision makers into reliable decision support systems

Decision-making is not a centralized activity in a number of large, artificial and natural enterprises, but rather a distributed effort of a multitude of agents. Typically, these agents are limited in,their view of, and influence on, the enterprise and are therefore restricted in their decision-making abilities. Despite these limitations, they can operate the enterprise satisfactorily by cooperating effectively within an organization of agents. Take, for instance, the scenario where the agents make correct, independent decisions more often than not and at the same rate. The assembly of these agents, through simple-majority vote, constitutes an organization whose decisions are more reliable than those of the underlying members. Simple-majority does not necessarily improve the quality of the overall decisions, however, if the agents display different probabilities of being correct. To this end, this paper proposes an alternative cooperation strategy to improve overall performance, as well as prevent the agents from being worse than the best member. The paper develops a theoretical framework to assemble reliable organizations, from lesser agents, and illustrates its potential in an application to face recognition with artificial neural networks. Significance: The size and complexity of problems found in industrial floors, such as production planning and scheduling, can make the decisions thereof intractable by a single, monolithic agent. To that end, this paper proposes a framework to synthesize effective decision-making organizations from less reliable agents.