An architecture for control and communications in distributed artificial intelligence systems

An architecture and implementation for a distributed artificial intelligence (DAI) system are presented, with emphasis given to the control and communication aspects. Problem solving by this system occurs as an iterative refinement of several mechanisms, including problem decomposition, kernel-subproblem solving, and result synthesis. In order for all related nodes to make optimum use of the information obtained from these problem-solving mechanisms, the system dynamically reconfigures itself, thereby improving its performance during operation. This approach offers the possibilities of increased real-time response, improved reliability and flexibility, and lower processing costs. A major component in the node architecture is a database of metaknowledge about the expertise of a node's own expert systems and those of the other processing nodes. This information is gradually accumulated during problem solving. Each node also has a dynamic-planning ability, which guides the problem-solving process in the most promising direction and a focus-control mechanism, which restricts the size of the explored solution space at the task level while reducing the communication bandwidths required. It also has a question-and-answer mechanism, which handles internode communications. Examples in the domain of digital-logic design are given to demonstrate the operation of the system.