This research investigates how artificial intelligence can be applied to the study of organizations. The authors construct a full artificial intelligence model of a small organization of cognitive agents communicating and cooperating to accomplish a task in which the interdependencies between agents can effect organizational performance. Each artificial agent is represented on its own computer using an extremely sophisticated software architecture, called Soar, that is capable of serving as a basis for general intelligence and can learn from experience. Concurrently with these studies, the researchers conduct studies in which human subjects participate in a "real world" version of the task. The task involves retrieving units at various locations from a warehouse. The research question under investigation is: "Given an organization of intelligent agents that are capable of learning, what coordination schemes are most effective and under what conditions?" The specific conditions examined are: the level of training, the type and amount of information shared, and the complexity of the task. A prototype version of the system, called Plural-Soar, has demonstrated the feasibility of this approach. The researchers intend to: 1) extend and refine the Plural-Soar modal to include all aspects of the task; 2) realize the model on multiple machines that can communicate; 3) conduct a series of human studies to be used to test, refine, and extend the set of Plural-Soar models; 4) compare the results of the simulation and human experiments; and 5) work toward the development of a computationally based theory of micro-organizational behavior.
