Multi-Agent Systems
Multi-agent systems study how multiple autonomous, interacting agents—each with its own information, goals, and decision-making—coordinate, cooperate, or compete to achieve individual or collective outcomes.
Definition
A multi-agent system is a system of multiple autonomous agents that interact within a shared environment, where each agent perceives, decides, and acts, and the outcome depends jointly on the agents' choices.
Scope
This area covers systems composed of many interacting agents and the principles governing their interaction: the game-theoretic analysis of strategic behavior and equilibria, coordination and cooperation among agents, distributed problem solving and constraint satisfaction, and mechanism design for engineering interactions with desirable properties. It treats agents as decision makers whose choices affect one another. The internal decision-theoretic machinery of a single agent is covered under reasoning under uncertainty, and learning in interaction belongs to the machine-learning subfield.
Sub-topics
Core questions
- How do self-interested agents behave strategically, and what stable outcomes (equilibria) result?
- How can agents coordinate their actions and cooperate toward shared or compatible goals?
- How can a problem be solved by distributing it among multiple agents with partial information?
- How can the rules of interaction be designed so that desirable system-wide outcomes emerge?
Key concepts
- autonomous agents
- strategic interaction and game theory
- Nash equilibrium
- coordination and negotiation
- cooperation and teamwork
- distributed problem solving
- mechanism design and auctions
- agent communication
Key theories
- Game-theoretic equilibrium analysis
- Game theory models agents as rational decision makers whose payoffs depend on others' actions, and equilibrium concepts such as the Nash equilibrium predict stable joint behavior, providing the analytical foundation for strategic interaction among agents.
- Coordination and cooperation
- Agents with limited information and overlapping goals must coordinate to avoid conflict and cooperate to achieve joint outcomes, using protocols, negotiation, and shared conventions studied in the multi-agent-systems literature.
- Mechanism design as inverse game theory
- Mechanism design engineers the rules of an interaction so that, even when agents act in their own interest, the resulting equilibrium achieves the designer's objective, such as efficiency or truthful behavior.
Clinical relevance
Multi-agent techniques are applied in automated trading and auctions, electronic markets, traffic and network routing, distributed sensor networks, robotics teams and swarms, supply chains, and the design of online platforms, wherever many decision makers interact and their incentives and coordination must be managed.
History
Multi-agent systems grew from distributed artificial intelligence in the 1980s, blending agent theory with game theory and economics. The 1990s and 2000s saw the consolidation of agent architectures, communication standards, and game-theoretic foundations, set out in texts by Wooldridge (2009) and Shoham and Leyton-Brown (2009), with mechanism design and auctions becoming a major applied thread.
Key figures
- Michael Wooldridge
- Yoav Shoham
- Kevin Leyton-Brown
- Nicholas R. Jennings
- Katia Sycara
Related topics
Seminal works
- shoham2009
- wooldridge2009
- jennings1998
Frequently asked questions
- How does a multi-agent system differ from a single intelligent agent?
- A single agent reasons and acts to achieve its own goals in an environment. In a multi-agent system, several autonomous agents act at once, so each agent's best choice depends on what the others do. This introduces strategic interaction, coordination, and incentive issues that single-agent AI does not face.
- Why is game theory central to multi-agent systems?
- Because outcomes in a multi-agent system depend on the joint actions of self-interested agents, game theory provides the tools to predict stable behavior (equilibria) and to design interactions. It lets researchers reason about competition, cooperation, and incentives in a principled way.