Connections with other fields
Multi-agent systems is a field spanning many disciplines of the sciences and the humanities. The study of social situations where many "agents" collaborate or compete, and the resulting dynamics, were since antiquity the object of political philosophy (agents = citizens, game rules = legislation, action sets = individual rights, etc.). The modern study of multi-agent systems is heavily influenced by economics, and in particular the Von Neumann-Morgenstern expected utility maximization principle in decision theory, as well as game theory, initiated by Nash's work in the early 1950s.Computer science and applied mathematics have in recent years explored the problems in multi-agent systems under a different eye, mainly focused on (i) algorithms and (ii) computational complexity. The algorithmic approach focuses on game dynamics and how communication, partial knowledge of the world, and modeling decisions alter the system's properties. The complexity branch mainly concerns itself with what is computationally feasible and how hard each particular problem is to solve.
Also important in recent years is the experimental aspect of multi-agent systems. Many times an optimal solution is very hard to compute. Also, when humans are part of the multi-agent system, their behavior only partly adheres to the strict utility maximization principle. Recent research aims to explore these grey areas, usually by making systems adaptive to different forms of behavior or providing better models for human choices. Important in this respect are the fields of behavioral economics and psychology.
Academic conferences
The main conference for multi-agent system is AAMAS (autonomous agents and multi-agent systems). You can visit the website for AAMAS 2008 (held in Portugal) here.Also important in the field is ACM's conference on electronic commerce (EC). Last year's EC (EC08) was held in Chicago, IL.
Research at Harvard
The Artificial Intelligence Research Group (AIRG) is very active in the field of multi-agent systems.Radhika Nagpal and her students develop multi-agent systems composed of very simple, limited capability agents, which, however, exhibit complex and nuanced properties on the whole. Avi Pfeffer uses probabilistic models to study decision-making in complex environments and also studies the reasoning patters of agents in games. Barbara Grosz and Stuart Shieber are famous for work in computational linguistics as well as human decision making, for which they have developed a rich experimental architecture (Colored Trails). Finally, David Parkes, the instructor for this course, has a very diverse research portfolio spanning mechanism design, stochastic optimization, game theory and social computing.