A Distributed Ai Architecture Enabling Multi-agent Cooperation

additional knowledge are automatically generated and integrated for the use of the cooperative community. As a consequence ISs are transformed into Cooperative Agents. In the consultation mode, the Shell provides the means for user interaction with either a separated agent or a set of agents (a society of agents) pursuing an overall common goal. This paper describes the general principles of cooperation within a community of coarse grain, semi-autonomous agents which are suitable for large industrial applications. The control mechanism that directs the activity inside each agent uses the blackboard paradigm to enable flexibility and integration of different modules. The way each agent cooperates is derived by domain independent knowledge sources. All the domain dependent knowledge needed for reasoning about cooperation is dynamically accessed during run-time. Agent's capabilities for multiagent Negotiation and Conflict Resolution are described, as well as the enabling functions and information structures in which these capabilities are supported. UPShell generates different Agents each one running as two separated Unix processes (not necessarily in the same machine), using sockets to communicate asynchronously among themselves (Fig. 1). UPShell is implemented in Prolog using YAP compiler under Unix and also encompasses capabilities to use X Windows and Unix interprocess communication facilities. Our system, while being, like MACE [2], domain independent, is aimed at real size applications, namely industrial problems. We have used UPShell to model higher-level parallelism (cooperative and distributed expert systems, coarse grained and loosely coupled). Nevertheless it could also be used to model lower-level parallelism (distributed production rules).