This extended abst ract gives a basic introduction to the 
aims of a new project, funded by the European Commission, on the experimental 
analysis of large multi-agent systems.
The project involves the University of Edinburgh (Anderson, Fourman, Robertson, 
Sannella, Vasconcelos, Walton), the Instute for Artificial Intelligence in 
Barcelona (Agusti, Sabater, Sierra) and the University of Liverpool (Parsons, 
Wooldridge).
Engineers working on large, distributed, multi-agent systems face a problem which
differs from conventional software engineering. They build software systems which
must coexist withother systems about which little may be known, yet we wish the 
overall behaviour of the population of systems to be predictable in certain ways 
depending on the domain of application. There is a loose analogy to the biosciences, 
where the scientific response to the problem of understanding population behaviour
has been to build mathematical models at various different levels of granularity 
of detail and use these to help form hypotheses about the driving forces in very
complex ecological systems. We are beginning to follow a similar path in analysing
multi-agent systems. Two technical prerequisites for solving this problem are a 
framework within which to design and run experiments on models of large agent 
systems and clear software engineering methods which allow the results of 
experimental analyses to be related to agent design choices. These currently do 
not exist. There are numerous agent deployment systems but no convincing systems 
for modelling agent populations and their evolution. There are numerous software 
engineering methods but none of these translate easily to agent design.
Providing a combination of analysis and design in this area must therefore be a 
ground-up exercise,drawing up on fragments of research from related areas.
Section 2 gives an overview of what we hope to achieve by the end of the project.