What if artificial intelligence could support students in truly understanding material, rather than just completing assignments for them? Educators are increasingly concerned about AI’s potential to disrupt learning, but some are exploring how it can be used as a tool to enhance education when applied thoughtfully. At Harvard, Professors Greg Kestin and Kelly Miller have been testing AI-powered tutor bots in their introductory course, Physical Sciences 2, during the fall 2023 semester. Their goal was not to replace human instruction, but to create a more personalized and engaging learning experience. The bot was custom-built using course content and designed to guide students through concepts by offering hints, visual aids, and practice problems—never giving direct answers. The approach followed a “flipped classroom” model: students first engaged with the AI tutor to grasp foundational ideas at their own pace, then brought questions to in-person class sessions. This shift allowed classroom time to focus on deeper discussion, collaboration, and problem-solving. The results were promising. Students in the AI-supported class reported higher levels of engagement and motivation. They appreciated the ability to ask questions anytime and receive immediate, non-judgmental feedback—something often missing in traditional lectures where students may hesitate to speak up. Miller emphasized that the key to success lies in designing AI to prompt critical thinking, not replace it. “In a traditional classroom, students often passively listen and don’t get real-time feedback on their understanding,” she said. “The AI tutor changes that by encouraging active learning.” Kestin warned that simply handing students access to tools like ChatGPT without guidance leads to worse outcomes. “When students use AI to think for them, they skip the mental work needed to learn,” he explained. “They don’t develop problem-solving skills.” Instead, the ideal role for AI is like a calculator—supporting students during tasks but not doing the thinking for them. The tutor bot helps generate practice problems, analyze data, and visualize complex concepts, empowering students to engage deeply with the material. Harvard has since expanded AI support across campus. Tools like HUbot and PingPong were launched by Harvard Information Technology, and custom tutor bots are now available in courses such as Financial Reporting and Control at Harvard Business School and Math 21A. Kestin and Miller’s research is ongoing, with a focus on identifying which types of interactions between students and AI bots lead to the strongest learning gains. Future studies will also examine long-term impacts, including retention and mastery over time. Their work highlights a growing consensus: AI isn’t inherently a threat to education. When integrated with proven teaching principles, it can become a powerful ally—helping students learn at their own pace, build confidence, and develop deeper understanding.