Development and evaluation of an intelligent tutoring system that can control levels of support

350903034 @Tomoo UNO

Abstract

It is pointed out that intelligent tutoring systems highly interacting with learners cause the Assistance Dilemma: when and how intelligent tutoring systems support learners[6]. In such a context, learning systems are required to equip a function to control levels of support. Scaffolding[10] is one of the valid approaches for controlling levels of sup- port. Scaffolding is to attempt to forward learning in a step-by-step manner. Learner is provided with adequate "Scaffolding" from expert during the early stage of learning, and this "Scaffolding" is gradually faded with the development of learning. So far, some intelligent tutoring systems that can control levels of support have been proposed[2, 9]. In this study, we develop an intelligent tutoring system that can multi-dimensionally control levels of scaffolding with multiple levels, and examine the Assistance Dilemma empirically. In our study, rst, we de ne Levels of Scaffolding (LOS). And we developed an intelli- gent tutoring system for natural deduction (ND) learning. Our system can control LOS in both "Selection" and "Application" of the inference rules used to deduct conclusion from premise. We also conducted two experiments to examine the relationship among LOS, learning behavior, and learning effects, utilizing the intelligent tutoring system as an experiment system. In the rst experiment, we investigated the impact of LOS of intelligent tutoring on learning behavior and learning effects, and discussed the Assistance Dilemma issue. In the second experiment, we observed participant's Help Seeking behavior, and we suggest that learner can use a learning system to some extent in rational manner.