Betty's Brain is a software environment created at Vanderbilt University by the Teachable Agents Group to help promote students' understanding of metacognitive skills and to reinforce river ecosystem knowledge as part of a science curriculum. It is a qualitative constraint system, using node-link diagrams to represent models for teaching complex scientific and engineering domains in high school.
The system focuses around a main character, Betty, who has asked the students to teach her about river ecosystems. In this way Betty's Brain diverges from a classic intelligent tutoring system (ITS) and adopts the learning by teaching (LBT) paradigm where computer agent interactions are focused around completing a primary task unrelated to the acquisition of domain content knowledge.
More recently, Betty's level of artificial intelligence has been largely modified to increase the interactivity with the students. Betty's task is to interact with students as a "good" learner, one who has self-regulatory skills, might. By incorporating feedback related to these self-regulatory skills we have shown that students are better able to perform in future learning tasks.
Current studies are focused on the 5th grade classroom with approximately 100 students. As well, as of July 2007, the system is being developed to integrate directly into classroom curriculum for the coming semester with included tools such as Front of the Class Betty, developed at Stanford University.
As of 2013 it has been used in many experiments to test the effectiveness of building and examining dynamic models for instruction in scientific domains. In several studies of Betty’s Brain by Biswas and collaborators, they trained students by having them create models of the oxygen cycle in a water-based ecosystem and then assessed them by having them create models of the nitrogen cycle in a land-based ecosystem. This is called a transfer test and it is a standard technique in learning experiments. In both activities, the systems were presented with resources and the modeling language was the qualitative diagram language built into the system. Experimental controls tested various hypotheses to begin to determine what worked and what did not. This is a powerful environment for beginning to understand what is effective about building simulations. Other useful systems for studying the effects of modelling for learning are IQON and Colab.
- Leelawong & Biswas, 2008 Designing learning by teaching agents: The Betty's Brain system. International Journal of Artificial Intelligence and Education, 18(3),181-208.
Basu, Satabdi, Kinnebrew, John S., Dickes, Amanda, Farris, Amy Voss, Sengupta, Pratim, Winger, Jaymes, & Biswas, Gautam. (2012). A science learning environment using a computational thinking approach. Paper presented at the Proceedings of the 20th International Conference on Computers in Education, Singapore.
Beek, Wouter, Bredeweg, B., & Lautour, Sander. (2011). Context-dependent help for the DynaLearn modelling and simulation workbench In G. Biswas (Ed.), Artificial Intelligence in Education (pp. 4200–4422). Berlin: Springer-Verlag. Biswas, Gautam, Jeong, H., Kinnebrew, John S., Sulcer, Brian, & Roscoe, Rod D. (in press, 2012).
Measuring self-regulated learning skills through social interactions in a teachable agent environment. Research and Practice in Technology Enhanced Learning. Biswas, Gautam, Leelawong, Krittaya, Schwartz, Daniel L., & Vye, N. J. (2005). Learning by teaching: