How AI Doubled Student Engagement in Harvard’s Physics Course

The intersection of AI and education continues to open new frontiers, as exemplified by a groundbreaking experiment at Harvard University. The study, led by Professor Gregory Kestin and Senior Lecturer Kelly Miller, explored the integration of a custom-designed AI tutor into the “Physical Sciences 2” (PS2) course, a physics class tailored for life sciences majors. The result? Student engagement doubled, and the AI tutor significantly outperformed traditional teaching methods in terms of learning outcomes. Here’s a closer look at how AI transformed the learning experience for physics students at Harvard.

The Experiment: AI vs. Active Learning

In the fall of 2023, nearly 200 students enrolled in the PS2 course participated in an Institutional Review Board-approved study that compared traditional active learning methods with AI-driven tutoring. Traditionally, active learning methods—where students engage in group discussions and problem-solving under the guidance of an instructor—are seen as superior to passive lecture formats. However, the AI tutor, custom-built on a framework leveraging generative AI, presented a different kind of engagement that exceeded expectations.

Students were divided into two groups, alternating between instructor-led active learning sessions and AI-tutored lessons. Both groups were exposed to identical content, with one key difference: while the active learning group interacted with human instructors, the other group worked through the material at home with the AI tutor. The AI was not simply ChatGPT but a system crafted to follow pedagogical best practices, such as managing cognitive load, encouraging a growth mindset, and offering personalized feedback.

Doubling Engagement and Learning Outcomes

The results were astonishing. Students working with the AI tutor reported not only higher levels of engagement and motivation but also learned twice as much content in less time compared to those in the active learning classroom. Pre- and post-test results revealed that students using the AI tutor showed over double the learning gains of their peers in the traditional classroom setting.

One reason for this success is the AI tutor’s ability to provide personalized feedback and adapt to each student’s learning pace. In a typical classroom, some students struggle to keep up, while others may find the content too slow or repetitive. The AI tutor alleviates this by responding in real-time to students’ questions and guiding them through problem-solving, creating an individualized learning environment for each student.

AI Tutors: A Complement, Not a Replacement

Despite the remarkable outcomes, Kestin and Miller are cautious about overstating AI’s role in education. They emphasized that AI tutors are not meant to replace human instructors but rather to complement in-person teaching by handling introductory material. This frees up valuable classroom time for higher-order activities like advanced problem-solving, project-based learning, and collaborative work.

Moreover, while the AI tutor enhances learning efficiency, the researchers stressed the importance of ensuring that AI does not “think” for students. Instead, it must be carefully designed to help students build their critical thinking skills, ensuring that they engage with the material deeply, rather than passively receiving answers.

Expanding AI Tutoring Beyond Physics

Encouraged by the success in the PS2 course, Harvard is already expanding the use of AI tutors. Mathematics instructor Eva Politou plans to implement a version of the AI tutor in her multivariable calculus course, where students will use it to generate and explore questions. Meanwhile, the Derek Bok Center for Teaching and Learning is collaborating with Harvard’s Information Technology team to integrate AI chatbots into other large introductory courses.

The AI tutoring system, initially developed by Kestin in response to the launch of ChatGPT, is now being adapted for a range of academic subjects. This flexibility makes AI tutoring an attractive tool for educators looking to offer personalized support across diverse fields.

The Future of AI in Education

The findings from the Harvard study offer a glimpse into the transformative potential of AI in education. By customizing AI tutors to fit specific course content and leveraging AI’s ability to offer personalized feedback at scale, educators can create learning environments that engage students more deeply and improve learning outcomes significantly.

While AI holds great promise, its future in education must be handled with care. Kestin and Miller caution against over-reliance on AI systems without proper pedagogical design. The goal is not to replace traditional teaching but to enhance it, creating an education system that is more inclusive, flexible, and accessible to students of all backgrounds and learning styles.

As AI continues to evolve, it may well become an essential component of modern education, offering students around the world access to personalized, high-quality learning experiences that were previously unimaginable.