People who are paying the most attention stop asking “should we adapt?” and start asking “how fast can we move?” at a certain point in any big change. In terms of education and AI, that time feels like right now. And it looks like researchers at the University of Cambridge are the only ones willing to fully ask the question that no one else is: what is the point of an exam if a chatbot can already pass the tests we use to see how much students have learned? It’s a really unsettling question. It’s not because the answer is clear that it’s not.
The Digital Education Futures Initiative at Cambridge has been working with teachers to figure out what AI means for how students learn, not just how teachers handle it. A peer-reviewed paper published in the British Journal of Educational Technology is just one example of the work that came out of that initiative. It goes far beyond the usual talk about AI tools and plagiarism policies. Professor Rupert Wegerif and researcher Dr. Imogen Casebourne say that generative AI is more than just a problem that needs to be solved. Based on their point of view, it shows flaws in traditional education that were already there.
“Dialogic” learning is at the heart of the idea. In this method, students talk through problems instead of memorizing answers. This is a science class about gravity. In a typical lesson, students learn Newton’s laws, write them down, and then show that they know them on a test. The Cambridge model starts with a question, like “Why do things fall?” and then they work through possible answers in groups. Finally, they run their ideas by an AI chatbot that can speak like Aristotle, Newton, or Einstein. It sounds almost like fun. The goal, though, is serious: to put students in the middle of a real intellectual conversation instead of just watching a whiteboard.
Cambridge’s plan seems more thorough than most EdTech ideas because it takes into account the places where AI actually causes problems, not just chances. The phrase “cognitive poison” used by Wegerif is very sharp. When a student who is having trouble can give their essay to ChatGPT without worrying about being caught, they have little reason to actually think about what they are writing. Most of the responses from institutions have been about finding ways to stop this, like making submission rules stricter and using tools to mark text that was written by AI. In response, Cambridge asks if the assignment was even important enough to keep safe in the first place.
It’s still not clear how fast any of this will get to real classrooms on a large scale. Still being worked on are tools like ModeratorBot, an AI that can join student group discussions, gently push quieter voices to the front, and find new ways to look into things. A different tool in this ecosystem called QReframer does something that seems almost counterintuitive: it doesn’t answer a student’s question; instead, it questions the assumptions that led to the question. That’s harder to teach even to humans who have done it before.
As I watch this project develop, I get the impression that Cambridge is doing something that most of the EdTech industry hasn’t: they’re treating students like thinkers who need to be challenged instead of users who need to be engaged. There’s more to the difference than meets the eye. AI education tools are mostly made to keep people’s attention and make things easier for them to understand. What Cambridge wants to do creates purposeful productive friction.
It’s a different story if schools have the will, the training budgets, or the flexibility in their curriculum to follow through. But the study is out there. The right questions are being asked. And while most schools are still debating whether to allow AI at all, Cambridge is already well past that stage, which is exactly what it was meant to do.
