Integrating organizational knowledge for dynamic scalable multiagent systems

We propose a model for rational multiagent reorganizations. Borrowing from organization theory, this approach models organizational structure by its natural dimensions (e.g. stratification, diversification, preservation, distribution). Agents choose reorganizations that maximize expected utility by quantifying the value of increasing or decreasing the organization's structure along its dimensions. Thus reorganization control is reduced explicitly to a search in n-dimensional space characterizing organizational structure. Utility of reorganizations is modeled using decision networks. Local and organization-wide measurements provide sufficient data to choose changes that benefit local and global goals. The first phase of evaluation has focused on scalability with promising results. The second phase involves testing the rationality of the reorganization control.