Revolutionizing AI in Education with Teachable AI
Chris MacLellan, Georgia Tech
Chris MacLellan is an assistant professor in the School of Interactive Computing at Georgia Tech where he runs Teachable AI Lab (TAIL) His lab aims to improve the understanding of how people teach and learn and to build AI systems that can teach and learn like people do. His recent work explores the development of computational models of human learning and how these models can support the development of effective learning technologies at scale. Chris received his PhD in Human-Computer Interaction from the Human Computer Interaction Institute at Carnegie Mellon University. Prior to his work at Georgia Tech, he was an assistant professor at Drexel University in the Information Science and Computer Science
Departments. He received the Faculty Research Excellence Award presented by Drexel University’s College of Computing Faculty Research, which recognizes recipients for their scholarly research and prolific academic and professional contributions to the University.
We are working towards creating a teachable AI technology that lets teachers build and modify intelligent tutoring systems to support their classes. The aim is to build an interactive AI agent that lets teachers author tutors through teaching rather than programming. To create a tutor for a new topic, the teacher provides the agent with examples of how to solve problems and corrects it problem solving on novel problems—similar to how a teacher would teach a human student. Once the agent learns how to do the task, it can then provide one-on-one instruction and coaching to students at scale.
We are currently building a platform called Apprentice Tutors that let teachers create their own tutors using this technology. We are currently focusing on deploying tutors into classrooms through Blackboard, and actively testing tutors to support college algebra classes at TCSG. This past year, more than 2000 TCSG students have had access to tutors through our platform.
While we are working on the foundational AI research about how to teach machines, we are also collaborating with teachers to build effective tutors. As students interact with the tutors, we use their data to assess their knowledge and to personalize tutor behavior to their current learning state. We also use this data to analyze class performance and assess student learning.
One of the challenges we face is building technologies that are accessible and personalized to teachers and students. We are working with teachers to build tools they can use to create personalized educational technologies for their classes. On the technical side, we’re trying to advance AI technologies to enable learning from teachers with just a few examples. Our goal is to teach agents with as few as 10 demonstrations so that they can teach students effectively. This is a foundational AI challenge that we are working to overcome.
