Chopra, Amit K. and Singh, Munindar P. (2015) Generalized commitment alignment. In: Proceedings of the 14th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2015) :. IFAAMAS, pp. 453-461.
Abstract
The interoperability of interacting components means that their expectations of each other remain in agreement. A commitment captures what one agent (its creditor) may expect from another agent (its debtor). Chopra and Singh (C&S) motivate commitment alignment as a meaning-based form of interoperation and show how to ensure alignment among agents despite asynchrony. Although C&S’s approach demonstrates the key strengths of relying on commitment semantics, it suffers from key shortcomings, which limit its applicability in practice. One, C&S do not model commitments properly, causing unacceptable interference between commitments in different transactions. Two, they require that the communication infrastructure guarantee first-in first-out (FIFO) delivery of messages for every agent-agent channel. Three, C&S guarantee alignment only in quiescent states (where no messages are in transit); however, such states may never obtain in enactments of real systems. Our approach retains and enhances C&S’s key strengths and avoids their shortcomings by providing a declarative semantics-based generalized treatment of alignment. Specifically, we (1) motivate a declarative notion of alignment relevant system states termed completeness; (2) prove that it coincides with alignment; and (3) provide the computations by which a system of agents provably progresses toward alignment assuming eventual delivery of messages.