Intelligent tutoring systems differ from ordinary educational software in that they contain explicit representations of both the target subject matter and the student’s knowledge, and that pedagogical decisions about which feedback to give, which practice problem to propose, etc. are based, in part, upon those knowledge representations. We argue that a suitable representation should satisfy two criteria that we call fidelity and efficiency. These criteria are developed with respect to constraints, a particular representation designed to satisfy both. A comparison between constraints and production rules with respect to these two criteria highlights the advantages of the constraint-based approach over the rule-based, model-tracing approach.
展开▼
