Defining and testing metaadaptable agents

In this paper we formally present a methodology that allows us to adapt the intelligence of the agents conforming a multiagent system. We perform such adaptation of the intelligence by means of genetics. We call the resulting environments metaadaptable systems, as they allow to adapt the adaptive capabilities of elements. So, we propose systems in which the capability of the agents to adapt themselves to the environment (via mechanisms producing some kind of intelligence) is itself adapted according to the changes of their own environment. Actually, finding the optimal intelligence mechanism and the optimal key parameters that govern it is not a trivial task, as they dramatically depend on the environment that produces the events that an agent has to use to infer its own conclusions. We denote by intelligence mechanism any strategy to produce behavior that could be considered intelligent. We show a simple but illustrative example of such a metaadaptable system, which can be easily extended to deal with more real situations as the construction of software systems based on intelligent mobile agents. This system has been fully implemented, and some interesting conclusions have been extracted from it.