Using and re-using agents in multi-agent learning environments

With the advancement of agent theories, languages and technologies it is now possible to adopt multi-agent architectures in the construction of intelligent learning environments. However, most of such architectures fail to capitalise on one of the major advantages of such multi-agent systems: the independence and reusability of its agents. To deal with such problem, in this paper we will present a framework for building open learning environments based on multi-agent architectures. The framework defines a set of agents with distinct types, which communicate in a subset of the KQML language. Such types characterise the agents' responsibilities in the multi-agent system and how they are ordered according to a simple taxonomy. To illustrate how this framework is used, we present two learning environments: MulPen and ErgoSys. These two environments were designed to train two different subjects: ergonomics and physics. Both environments share the same general agents: the PACF-Server and the learner modelling server. And both of them re-used Vincent (the pedagogical interface agent) in their training sessions.