Meta agency and individual power

Abstract

We are at the frontier of deploying real intelligent agents, able to build his own under- standing of the world and of the processes there going. How can these features be achieved? We follow the trail of individual power, the framing concept for the agent’s inner capabilities to achieve dominant goals. Our research lead us to distinguish external from internal factors in the agent’s behaviour: while the former depend on the relation the agent has with his environment (including his pairs) and ties in with social power, the later link up to the agent control and are rather less understood. We are focused on the integration of will and structure inside the agent control. Why these factors in particular? We believe they lay in a core set required to achieve general intelligence. Will manages the commitments of the agent. Therefore, it encapsulates and defines —if not explicitly, certainly implicitly— the individual character traces (eagerness, fearless, etc.) and local attitudes (persistency, urgency, etc.) in face of occurring events. Control structures (also known as “architectures”), on the other hand, can range from the very simple, reactive-like rules processing, to the very sophisticated, mental state based, systems. Here we step to the synergic integration of control structure and commitment manage- ment: An experiment where two teams of agents compete for resource gathering. While the reference team features the well known reactive control, with the test team we apply new ideas exploring meta agency for agent control and will for behavior design. This experiment allowed us to illustrate the integration and a virtual application of previous contributes. The meta agent and mental states based control proved to be able to define flexible, effective behaviours. The will, by centralising the commitments management, eased the implementation, debugging and interpretation of those behaviours.