Teaching teenagers to work with AI, not to hide from it
Schools are debating whether to ban AI. The more useful question is what students should be able to do that a chatbot cannot. Here is why building beats banning.
The wrong debate
In staff rooms across the country, the same argument plays out. One group wants to ban AI tools because students use them to write essays and solve homework. Another wants to embrace them wholesale. Both sides are looking at the same thing: a machine that produces answers.
The more useful question is a different one. If a chatbot can produce a decent answer to almost any school exercise in seconds, what should a student be able to do that a chatbot cannot? Ask that, and the debate about banning becomes less urgent than the question of what we are teaching for.
The skill worth teaching is not producing an answer. It is knowing what to ask, judging what comes back, and building something with it.
Learn by making something that works
The most reliable way to teach a skill is to have students use it for something real. In a programme built around this idea, a group of high school students does not write an essay about AI. They build a small tool: an assistant that answers questions about their school's timetable, a script that sorts their class's survey results, a simple app that helps a local shop track orders.
Making a working thing changes the relationship with the technology. A student who has had an AI-generated function fail in front of the class learns, more memorably than any lecture could, that its output has to be read, tested and understood. A student who has to explain to a real user why the tool is slow learns about trade-offs. Mistakes become part of the process rather than something to hide.
Ownership matters too. When the project is the student's own, they have a reason to understand each line. Copying becomes pointless, because copied work cannot be defended in the five-minute demo at the end.
What students should leave with
A good course in this area is not a list of tools, because the tools will change within a year. It should leave students with abilities that outlast any particular product.
- Framing a problem clearly enough that someone, or something, can help solve it.
- Writing precise instructions and noticing when an answer is subtly wrong.
- Checking claims against a second source, rather than trusting fluent text.
- Breaking a large task into small steps, and testing each step.
- Presenting what they built, and explaining its limits honestly.
Notice that most of these are not technical. They are thinking skills that happen to be practised through technology. A student who can do them will adapt to whatever tool arrives next.
Teachers are part of the answer
Programmes that go straight to students and bypass teachers rarely last. Teachers are the ones in the room every day, and they need to feel confident rather than threatened. The most effective approach brings them in as co-learners: a short, practical workshop where they use the same tools, try the same exercises and discuss how their own subject changes.
A maths teacher might use an AI to generate ten variations of a word problem in a minute, then spend the saved time on the students who are struggling. A literature teacher might ask students to critique an AI's reading of a poem. The aim is not to replace the teacher's judgement. It is to give it more room.
Where this leads
The students in these programmes are not all going to become engineers, and that is fine. The goal is a generation that is neither frightened of AI nor naively dependent on it, who can sit in front of a new tool and quickly work out what it is good for and where it will let them down.
For schools, a practical way to begin is small: one class, one project, one term, a clear demo at the end, and an honest conversation afterwards about what worked. Banning is easy and teaches nothing. Building is harder, and it is where the learning is.
Keep reading
The AI works. Nobody uses it. What went wrong?
The pilot was a success, the launch email went out, and three weeks later everyone is back on spreadsheets. The missing piece was never the technology.
Read articleHow to pick your first AI use case, before someone picks it for you
After a workshop, the whiteboard holds forty ideas for AI. Most will disappoint. A simple scoring method, and a few ground rules, helps you start with the one that will actually succeed.
Read article