International Workshop on Massively Multi-Agent Systems (2004.12. 10-11)

Participatory Design: From Agent-Based Social Simulations to Large Scale Multi-Agent Systems

Alexis Drogoul

Laboratory of Computer Sciences, Paris 6

Changes in an open, large scale MAS can partly occur because of its own mechanisms of evolution and partly because it will have to adapt to the pressure of their environment. This environment is likely to include humans at different stages : designers of the system, software engineers, and, of course, end users. Participatory methods aim at designing concrete means for all these users to influence and shape complex agent-based systems to their needs through processes of imitation, learning and training. These methods come from the domain of agent-based social simulation, where recent works on "participatory simulations" have opened a whole new field of research, allowing experts and non-experts to interactively define the agents' behaviors, mostly through role-playing games. The agents that populate the MMAS will be autonomous or semi-autonomous, but nevertheless be designed to play a twofold role : that of computational structures for implementing the MAS and that of "assistant" agents dedicated to an user or an expert. When provided with adequate learning capabilities, they may autonomously adjust or acquire their behavior, in situation, through repeated interactions with (or observations of) their "user".

The challenges raised by participatory methods are :

- Design of tools to seamlessly "embed" the users (or experts, or even designers) as "agents" within these MMAS;

- Design of tools to enable the users to influence, shape, or change the behaviors of the agents they interact with ;

- Development of machine learning techniques to enable the agents to learn and generalise behaviors through these interactions;

- Development of techniques to allow the agents to spread these "innovations" throughout the MMAS .