Robots in the Classroom

When most people picture AI in education, they imagine students using ChatGPT to do homework and write essays. But that’s only one facet of what this technology can do in the classroom. My work focuses on something different: having students learn by teaching AI agents. 

The idea builds on a long-established maxim: that the true mark of expertise is being able to explain a subject clearly to someone else. In 2019, I developed Curiosity Notebook, a web-based platform where students could select a learning task and teach a robot. I ran a study during a four-week after-school program at an elementary school in the Waterloo, Ontario, area. Students broke off into small groups and read information about rocks, then taught their robot to classify them as metamorphic, sedimentary or igneous. 

Related: AI Is Ruining My Education

All this work pre-dated the popular explosion of generative AI. Our system used simple, scripted dialogue. The robot would ask students questions to help it identify the rocks: does it have crystals? Does it have layers? Is it sparkly? The students could also direct the conversation and teach in their own way, rather than just responding to the robot’s questions.

My research showed that students were highly engaged using this educational method, mainly because of the social dynamic it cultivated. They weren’t learning alone; they were working together to teach the robot. The robot even nudged quieter students to speak more when it noticed others were dominating the conversation. 

Over the next year, my research will focus on leveraging generative AI to build more sophisticated versions of the Curiosity Notebook platform. I’d like to program robots to make mistakes intentionally to promote certain kinds of learning. Imagine a kid who’s studying programming—the AI could make simple coding errors at first, then more complicated ones over time, like a video game with progressively harder levels. I’d also like to design AI agents to respond thoughtfully to the personalities of the kids teaching them. We found that when a student who likes self-deprecating humour interacts with a self-deprecating agent, for instance, they feel more confident teaching the robot and enjoy the process more.

I’d love to see classrooms where AI is integrated as a core part of curricula to boost collaboration between students. AI agents could come in many forms, like mobile humanoid robots or discreet devices resembling a Google Home console. Or they could be something more creative, like an artificial flower or tree. This “smart” classroom plant could change colour based on the collective mood of the class, glowing red if most students struggled with an assignment, for instance. Whatever the form, the goal should be the same: to use AI to foster social connection and collaboration in the classroom, rather than isolating each student behind a screen.

Related: I Let an AI Avatar Teach My University Course

Looking ahead, one of the most promising applications of AI is in reflection, an exercise that encourages students to think critically about their learning experiences. Instead of having students write an essay in response to a prompt, an AI agent would ask follow-up questions about lessons and help students articulate complex thoughts. The conversation could be recorded and shared with teachers for assessment, while also giving students something to revisit later. Reflection could even take new forms, such as voice notes, drawings or other creative mediums, making the process more accessible and meaningful.

AI presents opportunities for teachers, too, reducing their workload so they can focus more on educating and mentoring. The tech could take care of routine, time-consuming tasks like grading assignments. With real-time updates on how students are progressing, teachers could also quickly see who needs help and when.

The possibilities of personalization are powerful. AI agents could be tailored to a student’s language skills, reading level and familiarity with the subject, which matters for neurodivergent children or immigrant kids still building literacy in English or French. Of course, it’s crucial to design guardrails on what an AI agent can say, since even small comments can shape how impressionable students see themselves.

The exciting thing about this work is that we’re moving beyond the narrative of AI as a harmful shortcut. Instead, we’re discovering how it can be a catalyst for deeper learning, collaboration and curiosity. Not by giving students answers, but by giving them the opportunity to teach.

Edith Law is a Google Research Chair in the Future of Work and Learning at the University of Waterloo.