Dr. Agneta Gulz and Dr. Magnus Haake present their research on the use of teachable agents in educational settings. A teachable agent is a computer character that is taught by a student. In this scenario, artificial intelligence techniques guide the agent's behavior. The agent makes decisions based on what it was taught in the past, and the student can revise the agent's knowledge (and her own) based on the agent's behavior. In this presentation, Gulz and Haake discuss the strengths and the limitations of the teachable agent approach, and present their research examining for which learners benefit most from the "learn by teaching" pedagogy.