Nature and importance of the subject

Multi-agent systems are set to be the key technology for software organisation during the next decade. While there have already been a number of multi-agent systems developed, the programming technology available for constructing such systems is relatively immature. Hence, there is a need for a powerful, general purpose programming technology for multi-agent systems.

The intention of this seminar is to bring together the leading researchers in these areas and to foster interaction between the various groups and thus get a better understanding of the ways in which multi-agent systems may be programmed in the future. As well as targeting logical approaches, a key element is to consider the requirements for efficient systems scaling within real world applications.

Over many years, work on computational logic has spawned research areas such as knowledge representation (KR), nonmonotonic reasoning (NMR), automated deduction (AR), and deductive databases (DDB). Each of these can be seen as an essential component within multi-agent systems, as agents need to

• describe the world (KR),
• reason somehow about how the world behaves (AR),
• decide in the light of uncertain information (NMR), and
• deal with massive data stored in heterogeneous formats (DDB).

In parallel, work within the multi-agent systems community has involved developing, often via logic, concepts concerned with communication languages and distributed computation (CC), cooperation and teamwork (TW), and the dynamic development of agent organisations (ORG). Again, each of these aspects can be seen as being required in complex multi-agent systems, as agents need to communicate with other distributed agents (CC), cooperate with other agents in order to achieve some goal (TW), and evolve , dynamically, organisational structures appropriate to the particular situation (ORG).

Goals of the Seminar

The seminar was set up in a way to allow ample time for discussions. We restricted the presentations to 30-35 minutes and allowed 10-15 minutes time for discussion after each presentation. This concept allowed for four talks in the morning and two talks after lunch.

We also set up four working groups: (1) Programming negotiation in agents, (2) Programming deliberation/rationality in agents, (3) Information/Data management via logic-based agents, Programming cooperation in agents. Participants had been allocated to these groups three weeks before the seminar started. Each working group was chaired by two senior researchers [with the exception of Working Group~4, where one of the co-chairs dropped out at the last moment] who contacted the participants and distributed material before the seminar. The groups met on Monday and Tuesday from 4-6 pm.

The idea behind these working groups was:

1. to identify key exemplars/problems that are relevant to that area;
2. to describe these exemplars/problems concisely/abstractly (can some of them be used as benchmarks/prototypical examples to check particular frameworks against?); and
3. to find out if, and to what extent, logic-based programming of multi-agent systems is useful for solving these problems.

Results were presented on Thursday, where all participants met from 4-6 pm.

An ambitious outcome that we aimed for was

A set of challenge problems/exemplars for logic-based programming of multi-agent systems. In addition, some criteria to determine whether a logic-based approach is useful or not. Or a list of problems where other methods are superior.

Outcomes of the Seminar

A homepage for the seminar has been set up, at http://www.cs.man.ac.uk/~zhangy/dagstuhl, containing all the presentations, the results of the working groups, and, last but not least, some photos of our official excursion: a wine tasting in Riol. As can be seen from the programme of presentations available on that web site, the seminar contained a wide variety of high-quality talks. Many participants commented on the excellent programme.

The working group idea generally worked well, with the groups often meeting outside their scheduled times. While the overall goal of the groups was perhaps too ambitious (after just two meetings), some interesting results have already emerged. We are currently trying to get the groups to continue their work (and, indeed, most seem keen) and hope that something useful and publishable will come out of it.

Following interactions during the seminar, it was decided to propose a new workshop on Languages, Tools and Techniques for Programming Multi-Agent Systems for AAMAS 2003 in Melbourne, Australia. This event is the most important conference on agent-based systems and is held annually. Over 12 seminar participants are now involved in the programme committee for this proposed workshop, and the time at Dagstuhl allowed us to work together on the application.

It has also been decided by several participants of the seminar to set up a steering committee for organising and continuing the CLIMA workshop series ( Computational Logic in Multi-Agent Systems ), which is closely related to the topic of the seminar.

Another important outcome of the seminar was to develop the details of a special issue of Annals of Mathematics and Artificial Intelligence on the topic of "Logic-Based Agent Implementation''. Again, interactions at the seminar led to the publication of the call for papers for this initiative; see http://www.csc.liv.ac.uk/~michael/LBAI03.