This paper presents an architecture for an agent capable of deliberation about the creation of new agents, and of actually creating a new agent in the multi-agent system, on the basis of this deliberation. After its creation the new agent participates fully in the running multi-agent system. The agent architecture is based on an existing generic agent model, and includes explicit formal conceptual representations of both structures of agents and (behavioural) properties of agents that can be used as requirements. Moreover, to support the deliberation process the agent has explicit knowledge of relations between structure and properties of agents. To actually create a new agent at run-time on the basis of the results of deliberation, the agent executes a creation action in the material world, which leads to a world state update to include the new agent, after which the new agent functions within the multi-agent system. This approach enables the design of evolution processes in societies of agents for which the evolution is not a merely material process which takes place in isolation from the mental worlds of the agents, but allows for interaction between mental and material processes. A combined mind-matter approach results in which the agents in a society can deliberatively influence the direction of the evolution, comparable to the potential offered by genetic engineering. The architecture has been designed using the compositional development method DESIRE, and has been tested in a prototype implementation. It is discussed how the approach introduced here can be used as a basis for automatic evolution of multi-agent systems for Electronic Commerce.

CWI
Software Engineering [SEN]

Brazier, F.M.T, Jonker, C.M, Treur, J, & Wijngaards, N.J.E. (1998). Deliberate evolution in multi-agent systems. Software Engineering [SEN]. CWI.