Operational modeling of agent autonomy: theoretical aspects and a formal language

Abstract. Autonomy has always been conceived as one of the defining attributes of intelligent agents. While the past years have seen considerable progress regard-ing theoretical aspects of autonomy, and while autonomy has been identified as an enabler for new computing paradigms such as grid computing, (web-)service-oriented computing or ubiquitous computing, autonomy as a software property is still miles away from implementation. The main reason for this is that while software can be given the autonomy to act on its own, it will always be the designers or users who are responsible for its actions in a legal sense. Hence, the implementation of autonomy will require rigorous modelling and verification, so as to ensure maximum dependability of systems acting autonomously. We take a first step in this direction by introducing a formal language ASL (Au-tonomy Specification Language) that allows for a precise specification of the ac-tivities to be carried out by a set of agents, the deontic constraints imposed on these activities, and the implications of activity execution on particular constraints (i.e., constraint dynamics). Agent autonomy is implicit in an ASL specification as the degrees of freedom left to the agents for the execution of activities. Further, ASL allows for the automatic detection and handling of norm conflicts, such that conflicts can either be resolved at design time or appropriate measures can be taken for their runtime settlement. The expressiveness and flexibility of ASL w.r.t. the specification of autonomy-related properties is illustrated in the context of an agent-based electronic trading platform.