Current research in Computer Supported Collaborative Learning (CSCL) is oriented towards a definition of systems able to sustain human collaboration through two main philosophies: mirroring and guiding systems [Jermann et al.2001].

The former attitude involves the detection of the user activity and the constitution of some forms of visual representation of this activity. Through this feedback offered by the system, a change of attitude of the user is expected and measured. The latter attitude, implies the existence of a cognitive theory which should relate the measured data with a model of collaboration. From here, the system should be able to address critical aspect of the interaction that need to be corrected to maximise the outcome of the user experience (i.e., [Barros and Verdejo2000]).

Both approaches present structural and actual limits. The former is biased by the way the user can understand the visual representation and if s/he possesses the ability to switch attitude using this information. The latter is strongly challenged by the lack of computer models able to compete with the complexity of human reasoning. Examples using complex AI approaches are still missing.

A third way is needed between these two approaches described above, which can propel the research on the cognitive models at first and then in the translations of these models into computer algorithms (see [Dillenbourg et al.1996]). This is one of the objective of my study. My idea is to use a light AI approach, where the intelligence is not relied to be completely in the system. On the contrary, my approach argument in distributing the intelligence in a system constituted by the users plus the computational media.