Learning with Virtual Agents: Competition and Cooperation in AMICO

This paper deals with the design and implementation of AMICO1 , a prototype to enhance mathematics learning by interactions between a student (or a dyad) and virtual agents (several co-learner companions and a mathematician). This work is part of a multidisciplinary project, the project LINGOT that aims to give math teachers tools to improve their teaching. We first present the foundations of the project: the pedagogical assumptions, the design approach and the different kinds of virtual agents we propose. Then we present our virtual companions who allow students and teachers to use different learning and teaching strategies. They can play several characters: more or less trouble maker, more or less correct in their answers, more or less algebra speakers. In AMICO we propose two scenarios of interactions between students and virtual agents: competition and cooperation. We sketch the software architecture we propose in order to implement a simple authoring tool. Teachers specify problems, figures, the lists of common answers for students and companions. Then AMICO generates automatically and dynamically the dialogues between students and companions according to the companions’ characters and students’ answers. First user testing are encouraging and give perspectives.