Can Neural Imaging Investigate Learning in an Educational Task?

2 Abstract A methodology is described for using fMRI brain imaging to study how students learn with an intelligent tutoring system. Students were imaged while learning an algebra isomorph. A cognitive model, which is capable of learning the material, identified 6 predefined brain regions as reflecting various aspects of the problem-solving and learning process. The model predicted that the learning would be mainly manifested as reduced activation in a lateral inferior prefrontal region reflecting increased fluency in retrieving relevant declarative information. This prediction was confirmed but there was also decreased activation in a predefined fusiform region suggesting increased fluency in parsing the problem representations. Exploratory analysis also found an anterior prefrontal region whose activity predicted individual differences in error rates. The cognitive model was modified to include an increase in visual fluency. A careful assessment was made of the models' ability to predict the time course of the BOLD response over 1-minute problem-solving episodes. 3 Neural imaging has contributed to the understanding of many forms of learning. Most of this research has used relatively simple laboratory tasks involving such things as word memory, perceptual priming, or simple skill acquisition. The question of interest in this paper is whether neural imaging can inform an understanding of the kind of learning that takes place in school. Studies of school learning have tended to use data sources such as final test performance that are both very summary in nature and much removed from the time of the learning. Imaging data from the performance of an actual school task might provide information that would be informative about the critical learning events. We created a learning environment that reproduced the essential aspects of a Cognitive Tutor, an intelligent computer-based instructional system that which has been shown to have success in teaching high school mathematics (Anderson et al., 1995; Koedinger et al., 1997). These are deployed in over 1000 schools nationwide and interact with over 300,000 students in the United States (Koedinger & Corbett, 2006; Ritter et al., 2007, in press). These tutors interact with students as they solve problems providing instruction on an as-needed basis. The simplified system in this paper was designed to enable use in an fMRI scanner and to interact with a simulation of a cognitive model. Figure 1a illustrates its interface. The student selects parts of these equations by pointing and clicking with a mouse. As illustrated in Figure 1a, the …