Technology Support for Complex Problem Solving: From SAD Environments to AI

Gautam Biswas, Daniel Schwartz, John Bransford, & The Teachable Agent Group at Vanderbilt (TAG-V)

Smart Machines in Education: The Coming Revolution in Education Technology, K.D. Forbus and P.J. Feltovich (eds.), AAAI/MIT Press, Menlo, Park, CA 2001

Abstract

For the past decade, our Cognition and Technology Group at Vanderbilt (CTGV) has been studying how technology can help students learn to approach the challenges involved in solving complex problems and learning about new topics. Work centered around our video-based Jasper Woodbury Problem Solving Series represents one example; a book written by our group summarizes this work (CTGV, 1997). A summary of work that goes beyond the Jasper series appears in CTGV (1998; in press).

We note in the Jasper book (CTGV, 1997) that our work began with simple, interactive videodisc technology, plus software for accessing relevant video segments on a “just-in-time” basis. We needed the interactivity because Jasper adventures are 20 minute video stories that end with complex challenges for students to solve. All the data relevant to the challenges (plus lots of irrelevant data that students have to sort through) have been embedded in the story line. An overview of the Jasper Series is illustrated in After viewing a Jasper adventure, students usually work in groups to solve the challenge (A CD that accompanies the Jasper book illustrates this process; see CTGV, 1997). To succeed, students need access to the data embedded in the Jasper story. Even if they cannot remember the details about required data, they can usually remember where in the story the data had been provided. The software lets them return to the relevant part almost instantly. For example, in the RBM adventure, students may return to the flying field scene to review how much fuel an ultralight contained; go to the restaurant scene to find a conversation that explained how large the landing field was in Boone’s Meadow; access Dr. Ramirez’s office to review how far it was from one city to the next, and so forth.

The software environments developed for each of the twelve Jasper adventures are very simple for students and teachers to use and extremely helpful as a support for student learning. From the perspective of AI, however, the software is trivial. Our approach to research has been to start with SAD environments (where SAD stands for “Stone Age Designs”) and to add sophistication and complexity only as necessary to achieve our instructional goals. Our SAD approach has allowed us to work closely with hundreds of teachers and students and, in the process, identify and test situations where increased technology support can further facilitate learning. In this chapter we especially emphasize situations where AI insights and techniques have become extremely helpful.

We describe two examples where principles from AI have allowed us to improve student learning. One involves creating an Adventure Player program, plus offshoots of that program, to accompany the Jasper Series (Crews et. al, 1997). A second involves the use of AI techniques to create “Teachable Agents” whom students explicitly teach to perform a variety of complex activities. (The emphasis on teachable agents is different from an emphasis on learning agents that learn on their own without the user explicitly teaching the agents, and assessing the adequacy of the agents’ new knowledge. In other words, our teachable agents do not have machine learning algorithms embedded into their reasoning processes.) Our work on teachable agents is quite new, so the ideas and data we present are still preliminary. We hope that our discussion of this project will help connect us with others who can provide insights about ways that we can strengthen and accelerate our current work.