Showing papers on "Multi-agent system published in 1988"

A knowledge-based model of distributed problem-solving

Abstract: The majority of artificial intelligence research has concentrated on applying knowledge-based programming techniques to program individual problem-solving agents with large amounts of knowledge thus enabling them to solve difficult problems. These agents work well with problems that are within the scope of their knowledge, but their performance degrades swiftly when they encounter problems that stretch past the boundaries of their expertise. A distributed problem-solving system consists of a collection of agents that are capable of performing knowledgeable, if not expert, problem solving and, more importantly, are capable of interacting to solve problems co-operatively. As a result, the agents are able to function as a team, resulting in a synergy which enables them to solve problems that could not be solved by any one individual agent. This dissertation proposes a knowledge-based model of distributed problem-solving which provides a conceptual framework for constructing co-operative problem-solving systems. In this framework, each agent is separated into two components: a problem-solving component which embodies the agent's own problem-solving knowledge and skill; and a planning component which embodies a knowledge-based model of the agent's own abilities and those of other agents in the environment and which uses this knowledge to plan and co-ordinate co-operative activities with other agents. The two components are tightly coupled, but each addresses a particular aspect of an agent's operation. The planning component determines when the agent should perform problem-solving functions for other agents and when the agent should have other agents perform problem-solving functions for it as well as which agents should perform this problem solving. The problem-solving component is only concerned with performing the tasks that the planning component determines should be carried out by the agent itself. The separation of an agent's distributed problem-solving functions from its basic problem-solving functions, coupled with a knowledge-based approach to representing each set of functions produces a coherent, extensible, and maintainable framework for facilitating co-operative problem-solving among groups of agents.