Just Enough Fidelity in Student and Expert Modeling for ITS

Abstract

Intelligent Tutoring Systems (ITSs) are usually comprised of three primary models – an expert model, a domain or system model, and a student model. Many of these models are quite complex to enable just about any learner to get the optimum tailored experience possible. These systems have shown great results, typically at least one standard deviation (a letter grade) better than traditional training (e.g. [1] [2]). This complexity not only ensured that ITSs were successful, it also prohibited their widespread use (e.g. [3]). Results of studies in which the expert and system models were simplified show similar gains in effectiveness (e.g. [4] [5]), suggesting that lower-cost ITSs can be just as effective as those developed at higher costs. This paper compares the results of effectiveness studies in which the ITSs had various levels of fidelity and presents some recommended guidelines in determining the level of fidelity for student, expert, and system models of the ITS.KeywordsCognitive ModelingPerceptionEmotion and InteractionMachine LearningNeural Networks Techniques for Data ProcessingAdaptive User InterfacesHuman performance improvementIntelligent TutoringAdaptive training