The influence of organizational hierarchies on the performance of distributed intelligent systems
Marcel Merx · 1989
Whenever several intelligent systems work together to reach a common goal, the question of selecting an appropriate organizational hierarchy arises. In this dissertation the organizational structures are generated by interconnecting the participating systems with three different types of links: communication, negotiation and control. Any two systems are connected to each other by only one of these links. Four systems were developed at the Center for Machine Intelligence at USC, each consisting of four agents and employing different modes of control to achieve a goal. The systems were subjected to a number of representative problems. An efficiency metric was defined to compare the performance of the individual systems for problems of different complexity. It was found that a system with four negotiating agents performed most efficiently in cases that converged to a solution. Although a system of centrally controlled agents converged to a solution for all test cases, it was found that this system was less efficient than the negotiating system. The test results are explained and suggestions for extensions of the USCCMI System given, as well as recommendations for future research.