Edmund H. Durfee and Young-pa So. "The Effects of Runtime Coordination Strategies Within Static Organizations." In Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI97), August 1997.
Young-pa So and Edmund H. Durfee. "Designing Organizations for Computational Agents," in K. Carley, L. Gasser, and M. Prietula (eds.) Computational Organization Theory, AAAI Press, 1997 (to appear).
Young-pa So and Edmund H. Durfee. "Designing Tree-Structured Organizations for Computational Agents." Computational and Mathematical Organization Theory, 2(3):219-246, Fall 1996.
Young-pa So and Edmund H. Durfee. Modeling and Designing Computational Organizations. In Working Notes of the AAAI Spring Symposium on Computational Organization Design , March 1994.
Young-pa So and Edmund H. Durfee. An Organizational Self-Design Model for Organizational Change. In Working Notes of the AAAI-93 Workshop on AI and Theories of Groups and Organizations , July 1993.
Young-pa So and Edmund H. Durfee. A Distributed Problem-Solving Infrastructure for Computer Network Management. International Journal of Intelligent and Cooperative Information Systems, 1(2):363-392, 1992.
Young-pa So and Edmund H.Durfee. Distributed Big Brother. In Proceedings of the Eighth IEEE Conference on Artificial Intelligence Applications, pages 295-301, March 1992.
We are developing a model of intelligent cooperation among communicating, autonomous, computational agents. The approach we take is called Organizational Self-Design, where agents freely initiate, change, join, leave, dissolve, and participate in organizations in order to jointly perform a set of tasks in an efficient, flexible, and reliable manner.
An organization is thought of as a long-term commitment made by the agents to a particular way of jointly handling the cooperative tasks. In the situations for which the organization was devised, the agents should cooperate effectively and reliably. However, many interesting real-world organizations are situated in complex, often uncertain, and changing worlds. If the collection of agents is to be effective in such circumstances, it must adapt to such changes in at least one of a number of ways. One way is to actively change the world such that the world once again matches the organizational structure of the agents. A second way is for the agents to reorganize themselves into an organizational structure that is better suited to the current situation. Finally, a third way is to adopt, at the outset, an organizational structure that provides sufficient flexibility to the agents such that they can dynamically adapt to new circumstances within the same organizational structure.
Clearly, any system of agents that will survive for an extended time should be capable of adapting in all of these ways. It should be able to perform task-environment (re)design (such as relaxing some goals or simplifying the environment) in cases where no organizational structure could possibly succeed, but when reorganization would permit the agents to accomplish their tasks in the environment, they should be capable of performing organization (re)design. Within a task-environment and an organization design, moreover, each agent within the system should be designed with sufficient local sophistication to be capable of choosing, from among its possible actions, the action that contributes most to the goals of the agent system.
These capabilities are all interrelated. As a specific example that has arisen in our work, the degree to which local sophistication needs to be designed into agents depends on the degree of flexibility allotted to the agents by the organization design. Or, conversely, the choice of organization design (in terms of flexibility given to the agents) should depend in part on the sophistication of the agents that are expected to populate the organization. Moreover, since, in many cases, the organization is being designed by the agents that will populate it rather than an external design specialist, we call the design process Organization Self-Design (OSD).
In our previous work, we have developed a model of OSD which models both the OSD process itself as well as various aspects of organizational performance. Our research has focused on providing a predictive performance model for a tree-like hierarchical organization for multiagent computational tasks. Such a model is necessary in order for the organization to evaluate alternative organization designs in the process of OSD, whether in a centralized, explicit form or in a distributed, implicit form.
In our ongoing work, we have been extending our model of OSD, along with our results in predictive analysis of organization performance in terms of system throughput and agent utilization. We have also been studying the interplay between the design of sophisticated agents and the design of organizations of such agents, and the impact of different design decisions on the flexibility and reliability of agent systems acting in dynamic environments.
The emphasis of this recent work is on questioning the traditional (distributed) artificial intelligence view that says that it is always beneficial to make agents more sophisticated (omniscient and omnipotent) and to make the protocols and coordination mechanisms among them richer and more powerful. While this is undoubtedly true in some circumstances, there are other circumstances where, at least when it comes to human organizations, populating an organization with unsophisticated agents is just as good as (if not better than) populating it with sophisticated agents. Our recent work has been examining this issue, concerned not so much with how to create sophisticated agents and powerful coordination mechanisms (which has been the traditional domain of (distributed) artificial intelligence), but rather with how the availabilty of sophisticated agents might change the organization design decisions and how different organizations demand the design of powerful coordination mechanisms to go along with them. Using both analytical and simulation tools, we can show the impacts of different combinations of organization designs and sophisiticated agents/coordination mechanisms.