Multi-Agent System (MAS), in AI, refers to a loosely coupled network of software agents that interact to solve problems which may be difficult or impossible for an individual agent or a monolithic system to solve. Each agent in the system is an autonomous AI entity that operates and makes decisions independently. Interaction protocols and strategies are defined that allow these independent entities to interact through cooperation, coordination, negotiation or competition. The true strength of MAS lies in the ability of agents to specialize and then combine their abilities to solve complex problems efficiently.
Historically, the concept of multi-agent systems began gaining traction in AI research in the 1970s. It was initially influenced by research in distributed computing systems and concurrency control in databases. The surge in interest in the 1990s was driven by the need to manage the complexity of large scale AI systems and handle decentralized information in an effective manner.
Some key contributors to the development of the concept of multi-agent systems include Marvin Minsky who ventured the "Society of Mind" theory proposing that intelligence could be a product of the interaction of non-intelligent parts. Moreover, Rodney Brooks with his subsumption architecture greatly influenced the design principles of MAS, emphasizing decentralized, emergent control in autonomous robots.