After our AI research: what do the exam results say?

After our AI research we had, among other things, an interview with Nederlands Dagblad. One of the questions they asked was so good that I'm still thinking about it 😊: if so many young people use AI, do we see that reflected in the exam results?

After that question we mainly shared our expectations, because... maybe AI pushes grades up, since pupils always have a personal tutor available, but it could also be that percentages actually drop, because pupils can partly skip learning.

So on this results day I dove into the numbers.

What struck me is that, since the rise of chatbots like ChatGPT, the exam results actually give surprisingly little reason for big claims. The percentages and grades have stayed reasonably stable in recent years. Only in 2023 do we see a clear dip, but that seems mainly linked to the end of the covid leniency measures.

Even more interesting to me was that the difference between school exams and national exams has barely changed either. You would expect exactly that difference if AI mainly helps pupils with homework and assignments outside the exam room. There too the percentages so far show no major shift.

We should bear in mind that national exams are normalised. Through the so-called N-term, differences in difficulty between exams are corrected for. That allows a fairer comparison between cohorts, but also means small changes in performance are not always immediately visible in the final exam grades.

Of course that doesn't mean AI has no influence on how pupils learn. But if that influence is there, it cannot yet easily be read from the classic exam statistics.