Protecting the Question: When Every Answer Is a Prompt Away

Protecting the Question: When Every Answer Is a Prompt Away 

On 28 September, I spoke in the first webinar of Teaching with AI, a three-part series for Fulbright Teacher Exchange alumni organised by the IREX. More than a hundred educators joined us from classrooms across the world, and the diversity of their contexts was striking. Some teach in schools where every child carries a device; others work in classrooms where electricity itself cannot always be taken for granted. Bringing people together from across the globe for a dialogue around what Davis Hassis called “the fire and electricity of our time” is possible only through the unifying efforts of IREX. I shared the session with two colleagues I have learned a great deal from—Dr Elizabeth Radday from Connecticut and Nanje Patrick Itarngoh from Cameroon. My task was to set the context for the conversation, and I decided to begin with a small incident from my own institution because it captured, for me, the strange place in which schools currently find themselves. I had spent almost fifty minutes with my students discussing how they could use AI as a thinking partner—how they could ask it to challenge an argument, offer another perspective, question their assumptions or help them identify gaps in their thinking. Ten minutes before the class ended, a colleague walked in to give instructions about an assignment and said, quite categorically, “Don’t use AI for preparing your assignment.” Every student turned towards me. I stayed silent. Not because I did not have an answer, but because that moment contained a much larger question. Within the same institution, sometimes within the same hour, we are telling children both to learn to use AI and not to use AI. Perhaps that contradiction is inevitable when a new technology enters an old system of learning.

I told the participants that we have had versions of this conversation before. We have spent decades worrying about calculators, for instance, and many mathematics classrooms still keep them away from children because we fear that students will stop learning to calculate. Yet the ability to calculate never automatically translated into the ability to think mathematically. We often confuse the mechanical act with the intellectual purpose behind it. The same distinction matters with AI. The question should not simply be whether a child has used AI; we should ask what the child has outsourced to it. If AI performs the mechanical layer—organising information, generating possibilities, checking language or offering examples—perhaps it can actually create more space for thinking. But if we allow it to take over curiosity, interpretation, judgement and meaning-making, then we have a very different problem. I find this distinction increasingly useful in my own teaching: what exactly are we outsourcing—the mechanical layer or the meaning-making layer? Banning a tool does not automatically protect thinking. Sometimes designing learning differently protects it better.

That led me to what I consider the more important question. AI can answer our questions, but it cannot ask the questions that emerge from our lived experiences in quite the same way. Authentic questions often come from wonder, confusion, discomfort, observation and struggle. They come when something does not make sense to us and we cannot leave it alone. So I shared some of the ways I have been experimenting with AI—not as an answer machine, but as a thinking partner. Ask the machine to question you instead of answering you. Ask it to argue against your position. Ask it to explain an idea from another person's perspective. Ask it to identify the gaps when you explain something back to it. I also shared two small examples from my own life. One involved a child cutting paper and a conversation that moved, almost unexpectedly, from paper to atoms and then to quarks. 

Elizabeth then moved the conversation from pedagogy to practice with her session, Build Your Own Assistant. She began with a question that made many of us smile: how many times this month have you told a chatbot that you teach ninth-grade biology? Her answer was the custom Gem—saved instructions that allow you to establish the context once instead of repeating it every time. She demonstrated a parent-communication assistant live and showed how easily teachers can build tools for repetitive work. But what stayed with me was the boundary she placed around the technology. Use AI for work that is repetitive, low-stakes and easy to verify. Keep your judgement for grading, sensitive situations and moments where the relationship itself is the work. That distinction becomes more important as these tools become more capable. Education contains a great deal of work that is inefficient, but it also contains work whose value lies precisely in the human attention we give it.

Nanje then brought a perspective that the global conversation on AI in education often misses. Drawing from his experience of teaching in very different classrooms, he reminded us that AI access and AI integration are not the same thing. A student does not necessarily need a device in their hand for AI to contribute to their learning. The teacher can work upstream—generate material, check it, contextualise it, adapt it and then bring it into the classroom through print, projection, a shared video or even a phone. Of course, that is a bridge rather than a destination, and we should not romanticise scarcity. But his point challenged the assumption that AI in education automatically means one device per child and an individual chatbot for every learner. We then moved into breakout rooms grouped according to the level of access participants actually had. I liked that design because the structure of the conversation itself acknowledged something we often forget: there is no single AI classroom. There are classrooms with abundance, classrooms with limitations and classrooms trying to negotiate everything in between.

I came away from the conversation with the same thought that has been occupying me for some time, but the webinar sharpened it further. The real challenge before education is not to decide whether AI is good or bad. That is a much easier question than the one we actually need to ask. We need to decide what we want to preserve when machines become extraordinarily good at producing answers. We need to protect curiosity, judgement, empathy, interpretation, disagreement and the ability to sit with uncertainty. We need to ensure that children do not lose the pleasure of struggling with an idea simply because an answer is available instantly. Perhaps this is the real pedagogical challenge of our time: not protecting children from AI and not surrendering children to AI, but using AI in ways that leave the intellectual ownership of learning with the child. When every answer is a prompt away, the classrooms that matter may be the ones that learn how to protect the question.