Open multi-agent systems

An open multi-agent system is a distributed system in which autonomous agents interact without centralized control, shared ontologies, or pre-established trust relationships. Unlike closed multi-agent systems — laboratory environments where all agents are designed by the same team with shared specifications — open systems are populated by agents built by different organizations, at different times, with different objectives and different assumptions. The defining challenge of open multi-agent systems is not coordination under shared assumptions but emergent coordination under heterogeneity: how do agents with no common language, no common ontology, and no common authority nevertheless produce coherent collective behavior? The failure of FIPA-ACL to achieve adoption in open systems, and the subsequent rise of natural-language-based agent communication via large language models, suggests that the field is moving away from formal specification and toward emergent, negotiated coordination. Open multi-agent systems are the natural habitat of agent economies, distributed systems, and the emerging ecosystem of AI agents that interact across organizational boundaries.

The design of open multi-agent systems is not a communication protocol problem. It is a social contract problem — and we do not yet know what contracts autonomous artificial agents can be expected to honor.