Implementing flexible and robust agent interactions using Distributed Commitment Machines

This paper focuses on the question of designing flexible and robust interactions between entities (such as agents or web services). Traditional approaches to interaction design focus on the permissible sequences of messages, but this constrains the autonomy of the entities, and limits interaction flexibility and robustness. The key to allowing for more flexible and robust interactions is to move away from message-centric design, and instead to design interactions using concepts such as the goals of the interaction, or the social commitments of the entities. This paper focuses on the Commitment Machine (CM) framework of Yolum & Singh, a simple and elegant approach for interactions which is based on social commitments. However, the CM framework is centralised, which is not a realistic model for distributed systems. The contribution of this paper is four-fold: an alternative formalisation of Commitment Machines is presented which is simpler than previous formalisations; Distributed Commitment Machines are defined as a more realistic model for interactions in distributed systems; the properties of Commitment Machines, both Distributed and centralised, are explored; and it is shown how the properties of Distributed Commitment Machines can be exploited to provide a simple implementation of CM-based interaction in distributed systems.