Pedagogical agents for learning by teaching: Teachable agents
Kristen Pilner Blair · 2007
We describe pedagogical agents called Teachable Agents (TAs) where students learn by teaching the computer. We describe an example of a TA, and discuss the features that allow students to capitalize on learning-by-teaching interactions. These include (i) explicit well-structured shared visual representations, (ii) independent performance of the agent, (iii) the agent’s ability to model productive learner behavior, and (iv) embedding the agent in environments that support teaching. Finally, we describe new directions for TAs including new models of homework practice, assessment, and video game environments. Almost everyone has had the experience of learning when teaching. Many graduate students observe they never really understood a topic until they had to teach it. To cultivate the benefits of learning by teaching, we have created a special kind of pedagogical agent that we call a Teachable Agent (TA). Students teach their TA and then assess its knowledge by asking it questions or by getting it to solve problems. The TA uses artificial intelligence techniques to generate answers based on what it was taught. Depending on the TA’s answer, students can revise their agents ’ knowledge (and their own). TAs do not replace real students. But, they do provide unique opportunities to optimize learning-by-teaching interactions. TAs, for example, always make their thinking visible, something that not all students can do. This raises the question: What aspects of learning-by-teaching can we maximize with TAs? We start with a concrete description of a TA. We then describe four core learning-by-teaching design principles that we believe can maximize learning-by-teaching. We conclude with instances of how these principles enable us to introduce exciting new technologies that leverage learning-by-teaching.