Abstract:

Intelligent tutoring systems are one of the many applications of artificial intelligence in education. They aim to simulate the activities of human tutors via the modelling of knowledge and savoir-faire (know-how) belonging to a specific learning domain, and the automation of their processing through the pedagogical expertise necessary to provide good teaching and support learners towards the acquisition of skills required for solving problems related to this learning domain.

To deal with the functional complexity of these environments, two conceptual approaches have emerged, each prioritizing only one of the main parameters of computerized tutoring: the density of the learning content and the accuracy of the monitoring of learning activities. The operationalization of these paradigms raises significant architectural issues, and their structural divergence induces a certain complementarity that recent works have highlighted while proposing a hybrid modelling approach allowing the compatibility of both knowledge tracing and content sequencing paradigms.

In this paper, we discuss upstream the approach that allowed us to propose model hybridization as a solution to the functional scalability issue of intelligent tutoring systems, and we propose downstream the architectural artefact on which this hybrid model rests on.

To achieve these objectives, we approach intelligent tutoring systems from a typical computer science perspective. Thus, as a feasibility study, a retrospective analysis of the literature and some popular systems allowed us to draw a critical appraisal of the conceptual, functional, and non-functional properties of intelligent tutorial systems such as definition, operational characteristics, architectures and tutoring models, effectiveness, efficiency, or usefulness. While explaining why we approach this problem as a software architecture flaw, we define what we mean by the term "functional scalability".

Confronting the results from our analysis phase with the constraints related to the principles of knowledge tracing and curriculum sequencing allowed us to propose the hybridization of both tutoring modelling paradigms as a solution to the problem of scaling up (tutoring in) intelligent tutoring systems. Other works explain how this can be achieved. The result on which this paper focuses is the presentation of the architectural artefact that supports this hybrid model.

The scaling up of artificial intelligence systems is a critical process that requires the collaboration of different knowledge representation formalisms and inference mechanisms. Thus, the mixed approach proposed in this paper, which aims to reconcile the conceptual logics of content sequencing and knowledge tracing approaches, relies on the compatibility and the flexibility between the rule-based reasoning mechanism, and the frame-based or object-oriented representation formalism (whose native mechanisms of instance back-propagation ensure the change of states and the processing business). We present some critical use cases of the operationalization of this architecture and discuss its advantages, limitations and perspectives.

Keywords:

Intelligent tutoring system, Adaptive learning, Software architecture, Artificial intelligence in education.