When I speak to schools and universities about AI, one conversation comes up again and again. Someone is convinced that a student has used ChatGPT to write an assignment. Someone else is equally convinced that the student wrote it themselves. And then begins the AI-tool battle to figure out who can detect AI text best. Except even parts of the US Constitution have apparently been flagged as AI-generated!
I’ve also had to pull out articles I wrote in the 90s to prove to a dismissive friend that yes, this was how I wrote back then. And this is where the entire idea starts becoming a little absurd.
AI is not that all-knowing English teacher who can take one look at a piece of writing and declare, “You clearly didn’t write this.”Generative AI itself has been trained on enormous volumes of human writing. Good writing, bad writing, formal writing, literary writing, academic writing. So we are now asking another tool to tell us whether our writing sounds too much like the machine that learnt from us.
Denmark has announced plans for something almost obvious: simply ask students to explain what they have written.
In August, its education ministry announced plans to add an oral defence to a major take-home written assignment on its HF upper-secondary programme. The details are to be worked out with school organisations. Schools are also encouraged to use screen monitoring during written exams and have students do more writing at school, where teachers can follow the work as it happens.
Essentially, Denmark is looking at changing the assessment itself. Handing in the finished piece would no longer be enough. The student would also have to sit there and defend it.Years ago in India, we found the idea of open-book examinations at American universities quite hilarious. How was it an exam if you were allowed to carry the book in? Surely all the answers were right there?
Except, of course, that was exactly the point. Having the book did not mean you knew what to do with it. You still had to work out which information mattered, how to apply it and whether it actually answered the question. If you could answer by copying a paragraph from the textbook, perhaps the problem was the question.
The book had not made the examination pointless. It had forced a clearer distinction between remembering something and understanding it.
AI is forcing us into a similar rethink. Only this time, the tool can do more than give us information. It can produce the answer itself. An essay, an argument, a conclusion - whatever is required. All the things we have got used to treating as evidence that someone has understood the subject.
So we have to ask what the student can explain, apply and defend once that answer is on the page. Can they spot where it goes wrong? Can they tell us why the evidence supports the conclusion? Can they use the same idea when the question changes?
We once wondered how an examination could work if students had the book. Now we are wondering how it can work if students have AI. And once again, we may have to think harder about the purpose of the examination itself.
India has now announced plans to train almost one crore young people in AI. We are clearly willing to put enormous resources into teaching young people how to use these increasingly powerful tools. But AI literacy without assessment reform creates a very strange contradiction. We encourage students to use AI to learn, research and produce work, and then assess them through tasks that these same tools can do extremely well.
So perhaps the question now is: what should always remain the student’s job?
Maybe writing complete, grammatically perfect sentences without an AI assistant will matter less in some assignments. But reasoning, judgment, knowing whether a source is credible, being able to defend a position and showing that you actually understand what you are claiming to know, those still have to belong to the student.
And perhaps this is the slightly uncomfortable thing AI is forcing us to confront. It has made cheating easier, certainly. But it has also exposed an old weakness in how we assess learning. A finished assignment could show us what a student had produced. We were too willing to assume it also showed us what they understood.
If AI can produce the answer, then the job of assessment is no longer to prove who typed the words, but to prove who owns the thinking.
