The Inside Story Behind Khan Academy's AI Tutor Khanmigo
Nov 19, 2023
auto_awesome
Kristen DiCerbo, Chief Learning Officer at Khan Academy, discusses the innovative AI tutor, Khanmigo, which uses GPT-4 technology. She shares her journey melding education with tech, highlighting the transformative potential for special education. Kristen explains the principles of mastery learning that guide Khan Academy’s approach, emphasizing personalized learning paths. She also reveals insights from collaborating with OpenAI and explores the future of AI in education, illustrating how technology can enhance student engagement and knowledge retention.
Khan Academy's Khanmigo utilizes GPT-4 to create engaging, interactive learning experiences that support mastery learning principles.
Kristen DiCerbo emphasizes the need for technology to address teacher shortages, highlighting the importance of digital environments in education.
Deep dives
Kristen DeCerbo’s Background and Perspective
Kristen DeCerbo, the Chief Learning Officer at Khan Academy, shares her diverse background in education and technology, starting as a school psychologist in a challenging bilingual environment. She emphasizes the role of technology in addressing the teacher shortage, especially in special education, which inspired her to explore digital learning environments. Her career journey led her from Cisco, where she contributed to networking academies, to Pearson, where she integrated learning sciences into broader educational content. This extensive experience has shaped her approach to scaling quality learning across various contexts, particularly in her current role at Khan Academy.
Khan Academy's Scale and Learning Philosophy
Khan Academy operates at a significant scale, with about 11 million monthly visitors and availability in around 50 languages, fostering a diverse learning environment. The organization's core philosophy revolves around mastery learning, where learners are encouraged to thoroughly understand a concept before progressing to the next. Research indicates that committing two hours per month on the platform yields significant learning outcomes, with a targeted one million learners achieving this consistent engagement. This mastery approach ensures students build a solid foundation, enhancing their ability to tackle increasingly complex skills.
The Introduction of Khan Migo
Khan Migo, an AI-powered tutoring assistant, aims to facilitate personalized learning experiences while ensuring students engage with content effectively. The development of Khan Migo was motivated by advancements in large language models, specifically GPT-4, leading to an interactive and supportive tutoring environment. Unlike standard AI interfaces, Khan Migo emphasizes guiding students through problem-solving rather than providing direct answers, adhering to pedagogical practices that prioritize student engagement. This tool employs targeted prompts and includes built-in safety measures, allowing learners to interact with content while maintaining a secure learning environment.
The Foundations of Learning Science in Khan Migo
Khan Migo is grounded in established learning science principles, particularly those that define effective tutoring methods. It provides immediate, stepwise feedback to students, addressing their specific mistakes and guiding them back on track, a capability that traditional educational tools often lack. The system also embraces the ICAP framework, fostering interactive learning; students engage in a dialogue rather than passive learning. The emphasis on engagement and dialogue sets Khan Migo apart, allowing for a more personalized and dynamic interaction that can enhance student understanding and retention.
Join two former teachers - Libby Hills from the Jacobs Foundation and AI researcher Owen Henkel - for the Ed-Technical podcast series about AI in education. Each episode, Libby and Owen will ask experts to help educators sift the useful insights from the AI hype. They’ll be asking questions like - how does this actually help students and teachers? What do we actually know about this technology, and what’s just speculation? And (importantly!) when we say AI, what are we actually talking about?
In the third episode of this new series, Libby and Owen talk to Kristen DiCerbo, the Chief Learning Officer at Khan Academy. Khan Academy is best known for its high quality instructional videos, often led by founder Sal Khan himself. It is used by millions of students around the world.
In March this year, Khan Academy launched Khanmigo, one of the first AI-based tutor and teaching assistants to use GPT-4 (one of the most powerful large language models). Kristen talks about what it was like to get a call from OpenAI and an early preview of GPT-4 last year. She explains how they built Khanmigo (spoiler – it’s not as complicated as you might think!), how it embeds principles from the learning sciences, and where she thinks this technology is going to lead us over the next few years.
Join us for this exciting episode as we speak to one of the most experienced and inspiring innovators in the field about how to actually build an AI chatbot for education!