Extending Multi-agent Cooperation by Overhearing

Much cooperation among humans happens following a common pattern: by chance or deliberately, a person overhears a conversation between two or more parties and steps in to help, for instance by suggesting answers to questions, by volunteering to perform actions, by making observations or adding information. We describe an abstract architecture to support a similar pattern in societies of artificial agents. Our architecture involves pairs of so-called service agents (or services) engaged in some tasks, and unlimited number of suggestive agents (or suggesters). The latter have an understanding of the work behaviours of the former through a publicly available model, and are able to observe the messages they exchange. Depending on their own objectives, the understanding they have available, and the observed communication, the suggesters try to cooperate with the services, by initiating assisting actions, and by sending suggestions to the services. These in effect may induce a change in services behaviour. Our architecture has been applied in a few industrial and research projects; a simple demonstrator, implemented by means of a BDI toolkit, JACK Intelligent Agents, is discussed in detail.