We consider the social relationships between autonomous principals, assuming that each principal is represented in a computational multiagent system via a unique agent. Work on commitment protocols has shown that the meaning of interactions may be expressed in terms of (social) commitments: C~p(x, y, p, q) represents an expectation of y (creditor) that x (the debtor) will bring about q (consequent) if p (the antecedent) holds. Further, we consider a decentralized setting where only the local states of the participating principals are available. Each local state reflects what the corresponding principal has observed, and carries the intuition of being a local projection of a notional global state. This raises the question of alignment: given that each agent is confined to its local view, but the commitments are interagent artifacts, how can we ensure that the agents remain in a coherent social state? For instance, when delivery relies on a courier, it will not be directly observable by the shopkeeper. The commitment to the delivery of the goods to the client, taken by the shopkeeper, is satisfied when the courier completes delivery. The client will see such a commitment as satisfied. Absurdly, the shopkeeper, who is not situated to observe this event, will not be able to do so.
展开▼
