Affective Meta Tutor

Project goal

The Affective Meta-Tutoring system is comprised of (1) a tutor that teaches system dynamics modeling, (2) a meta-tutor that teaches good strategies for learning how to model from the tutor, and (3) an affective learning companion that encourages students to use the learning strategy that the meta-tutor teaches. The affective learning companion’s messages are selected by using physiological sensors and log data to determine the student’s affective state. Its efficacy (and our hypotheses) will be tested by seeing if it creates lasting changes in the students' modeling practices when compared to the unaugmented modeling tool. Evaluations compared the learning gains of three conditions: the tutor alone, the tutor plus meta-tutor and the tutor, meta-tutor and affective learning companion.

Background and motivation

Research problem. In science and mathematics education, a well-known problem is that students often understand and manipulate scientific representations (models) in a shallow manner (Stratford,1997). For instance, when drawing causal networks, students sometimes pay little attention to what the nodes and links denote (Biswas et al., 2005). For instance, they might copy and paste a node's name from the text or change a link's label from + to - without considering what these edits say about causation. Although this is a well-known problem, it has no standard name, so let us call it the shallow modeling problem.

Hypothesized solution. When the modeling is done with a software tool, then it is becoming possible to detect episodes of shallow modeling from patterns of usage. Currently, this capability has been used for meta-tutoring. That is, when the meta-tutor detects shallow modeling, it gently reminds students to do deeper modeling and explains how, if asked. This is called meta-tutoring because it does not tutor the scientific domain (e.g., stream ecology) but it does tutor meta-cognitive practices for learning the scientific domain. Unfortunately, current work also indicates that when the meta-tutor is removed, students resume shallow modeling (Schwartz, Chase, Chin et al., in press; Roll, Aleven, McLaren et al., 2006). In line with current theories (e.g., Dweck's; Picard's), we hypothesize that lasting benefits require first changing students' cost-benefit beliefs about shallow vs. deep modeling practices, and then breaking their old modeling habits and instilling new ones. Moreover, these changes are more easily accomplished in a supportive social context. Thus, our solution is to combine meta-tutoring technology with the technology of affective learning companions (e.g., Bickmore & Picard, 2005), which have been used successfully to get people to make persistent changes such as adopting safer sexual practices (Read et al., 2006) or persevering in the face of frustration (Burleson & Picard, 2007).

Research plan. In order to generalize from earlier work, we will use system dynamics diagrams, which are a more sophisticated modeling language than causal networks. We will construct modules that use a systems dynamics modeling tool to teach interesting science to high school students during summer camps. We will collect log data, verbal protocols and non-invasive affect sensor data. These will be used to develop and calibrate a combined meta-tutor and affective learning companion. Its efficacy (and our hypotheses) will be tested by seeing if it creates lasting changes in the students' modeling practices when compared to the unaugmented modeling tool.