Date: 10/28 1:18 AM From: Robert H Guttman, guttman@media.mit.edu To: Distributed Systems, dist-sys-97@media.mit.edu Subject: Int'l Workshop on Multi-Agent Systems Int'l Workshop on Multi-Agent Systems ------------------------------------- Oct 12-14, 1997 MIT Endicott House, Dedham, MA This was an invited workshop with most of the top researchers in the field of multi-agent systems (MAS). Surprisingly, the workshop was largely about understanding who the MAS community is and what it's all about. The group wrestled with fundamental questions such as: What is our foundation? What is coordination in MAS? Are we a theoretical or an applied research group? Is there a clear framework and roadmap for our research and, if so, where are we today and where are we going? To help answer these questions, the group was broken up into smaller groups: 1. Foundations and Semantics of Multi-agent Systems including Multi-Agent Aspects of Affect, Beliefs, Commitments, Desires, Goals, and Intentions cochaired by Castelfranchi and Hewitt 2. Implementing Multi-agent Systems including Languages, Frameworks, and their Standards chaired by Decker 3. Market-Oriented Approaches chaired by Wellman 4. Multi-agent Applications including those situated in the "Real World" and that make use of Hardware cochaired by Hogg and Jennings 5. Multi-agent Assessing, Planning, and Predicting making use of Search, and Exploration chaired by Durfee 6. Multi-agent Learning and Teaching chaired by Sen 7. Multi-agent Collaboration, Coordination, Competition, and Negotiation chaired by Lesser 8. Social Aspects of Multi-agent Systems including Teams, Organizations, and Communities cochaired by Huhns and Tambe I have presentations for each of these groups if anyone wants to see them. Supposedly, a web site will host them all, but I don't think it's up yet. As probably expected, most of the questions above were not concisely answered. There was much talk about coming up with both a common architecture to share with the community so people could build upon each other's work (as opposed to working in isolation which seems to be the current state of affairs) as well as a common goal so people can evaluate their progress against others' efforts. The RoboCup competition came up as an example; however, it was considered insufficient for the purpose of a common goal/target because it's too much of a toy problem and it ignores many of the core issues in MAS research. Although people were looking for cohesion, I noticed that there were three general camps of researchers: 1. BDI & Sociological (e.g., mostly the Europeans) 2. Market-Oriented (e.g., Wellman, Sandholm) 3. Applied (e.g., Hogg) The BDI vs. MO split can be thought of as being the division across the cooperative vs. competitive boundary respectively. However, not everyone agreed with this. In fact, there were a lot of definitions and concepts that we couldn't get consensus on. This tended to cause a noticeable rift between the market-oriented crew and everyone else. I think this mostly stemmed from the MO group having a strong mathematical foundation for their work where everyone else was still trying to find their foundation. Wellman insisting that virtually all coordination can be modeled as MO resource allocation problem using currency didn't help matters. In general, the MAS community sees itself as a relatively young and small research group. I found them very friendly and welcoming and had a few terrific one-on-one talks (e.g., with Les Gasser, Katia Sycara, Tim Finin, and Tuomas Sandholm). They're wrestling with some tough issues and it's not clear if these will be resolved. I tried to warn them that they only have a short window of opportunity to get their ideas into the mainstream since so-called multi-agent system frameworks are emerging that are becoming popular (and are thus defining multi-agents systems) yet bear little resemblance to the work done in this community. I don't think I told them anything they didn't already know. - Rob PS: I have some more specific notes about this and that if anyone wants to know more.