One of the things that is lacking in the relatively young field of Artificial Intelligence and Education is a good framework for describing and designing Intelligent Tutoring Systems (ITS). From literature, an agreed upon architecture seems to emerge, consisting of three or four experts: A Domain Expert, a Student Expert, a Teaching Expert, and often an Interface or Presentation Expert. This framework works well for describing existing systems in general terms, but as soon as one dives somewhat deeper or wants to use it for designing a new system, problems arise. The forms of expertise turn out to overlap, and the interfaces between them are unclear. In this paper I describe a new conceptual framework for considering ITSs, consisting of three levels of problem solving loops: (1) A Curriculum level, (2) a Task level, and (3) a Discourse level. At each level potentially three processes take place: the Planning of actions to achieve some goal, the Monitoring of the execution of these actions (the performance), and the Diagnosing of possible errors in the execution. After a global description, I illustrate it by using it to describe two systems I have been involved in: the EUROHELP system, which actually is a shell for building Intelligent Help Systems for interactive computer programs; and the FysioDisc system, a coaching system for physiotherapeutic diagnosis. One of the advantages of the framework and the existence of well described interfaces between levels has already been demonstrated clearly by the porting of the entire communication level of the Eurohelp system to the FysioDisc system.
Winkels, R. (1990). A new framework for describing and designing intelligent tutoring systems. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 451 LNAI, pp. 230–243). Springer Verlag. https://doi.org/10.1007/3-540-52952-7_29