Multiagent Systems

What is an agent? A typical definition is that it is a computer system situated in some environment, and is capable of autonomous action in this environment in order to meet its design objectives. Often functionality and productivity may be increased by letting several agents interact with each other forming a Multi-Agent System (MAS). A MAS is often composed of agents of different kinds that have different roles. Multi-Agent Systems are often seen as a natural metaphor for conceptualising and building a wide range of complex computer systems

From a software engineering perspective agents can be seen as a natural extension of the concept of software objects by adding a new level of abstraction. Object-oriented programming added entities that have persistent local states, i.e., objects, to the structured programming paradigm. Similarly, agent-oriented programming adds entities, i.e., agents, that have an independent execution thread and are pro-active. An agent is able to act in a goal-directed fashion rather than just passively reacting to procedure calls.

What makes agents particular interesting from an AI perspective is that they may also be able to:

communicate with other agents and with humans,
learn from experience,
adapt to changes in the environment,
make plans,
reason using, e.g., logic or game theory,
move between different computers, and/or
negotiate with other agents.

Which of these abilities that are implemented in a particular agent depend on its tasks.

For questions related to the multiagent systems area contact Paul Davidsson, Blekinge Institute of Technology, phone 0457-385841 email paul.davidsson _at_ bth.se