I argue that the notion of "autonomy" in Agents - which can be even definitory - and in particular of "social autonomy" (autonomy relatively to other agents) should be defined and operationalised in terms of dependency relationships. I generalise our original theory of dependence relations - where dependence is relative to some lacking resource or ability - to include deontic conditions (permission, authorisation) and also information, instructions, control, etc. This view of autonomy, and this generalised notion of "resource" are strictly based on the theory of action and of its necessary inputs and powers, and in particular on the architecture of cognitive agents (specifically BDI ones). I claim that this framework gives a principled perspective to the theory of autonomy and predicts the various possible dimensions of dependence and regulation.
Keywords: Autonomous Agents, Philosophical foundations, Multi-Agent Systems, Cognitive Modelling
Citation: Cristiano Castelfranchi: Founding Agents' "Autonomy" on Dependence Theory. In W.Horn (ed.): ECAI2000, Proceedings of the 14th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2000, pp.353-357.