Control and behavior of a massive multi-agent system

Massive multi-agent systems are systems built with several thousands of agents. We characterize these systems as unsteady, evolutionary and adaptive. We show that their design is made while using an incremental agentification method, using functionalities distributed in the behavior of the agent organization. But modifications of the number of agents, modification of their structures and of their capacities of communication during the running is no compatible with an a priori complete specification nor with a fine prediction of the behavior. The control of this system achieves itself while endowing agents in their running with characters of self-assessment of their organization. We claim that we can control such a system while defining a new organization-of agents observing the initial in real time, valuing modifications of its morphological characters and using this assessment to modify the initial organization, in a systemic loop of self-control.