Multi-agent systems

• Authors: Victor R. Lesser
• Venue: Encyclopedia of Computer Science
• Year: 2003

Multi-agent systems are computational systems in which several artificial "agents", which are programs, interact or work together over a communications network to perform some set of tasks jointly or to satisfy some set of goals. These systems may consist of homogeneous or heterogeneous agents. Examples of agents would be ones for detecting and diagnosing network problems occurring on a segment of a local area network; for scheduling the activities of a group of machines in a workcell on a factory floor; or for locating agents that are selling a specific product and deciding on what price to pay. Agents may be characterized by whether they are benevolent (cooperative) or self-interested. Cooperative agents work toward achieving a set of shared goals, whereas self-interested agents have distinct goals but may still interact to further their own goals. For example, in a manufacturing setting, where agents are responsible for scheduling different aspects of the manufacturing process, agents in the same manufacturing company would behave in a cooperative way, while agents representing two separate companies, where one company was outsourcing part of its manufacturing process to the other company, would behave in a self-interested way. Agents often need to be semi-autonomous and highly adaptive due to their "open" operating environments, where the configuration and capabilities of other agents and network resources change dynamically. Agent autonomy relates to an agent's ability to make its own decisions about what activities to do, when to do them, and to whom information should be communicated. Scientific research and practice in this area, which is also called Distributed Artificial Intelligence (DAI), focuses on the development of computational principles and models for constructing, describing, and analyzing the patterns of interaction and coordination in both large and small agent societies.