Education · Idee
Are we still testing the right thing?
When the machine finishes the homework in twelve seconds, the homework no longer tests anything. That is not a crisis for schools — it is an overdue question.
In four sentences
- Exams measure results because results were long a good indicator of understanding. That link is broken.
- Bans don't solve this: they shift the problem to the time before submission.
- Formats that keep working test the path — interim stages, decisions, defense, oral follow-up questions.
- That is more expensive to run and more honest in its results.
The broken link
For decades, a finished essay was solid evidence that someone had worked through a topic — not because the paper counts, but because the way there ran through your own head. Exactly this link is gone. The result can now be produced without walking the path.
The usual reactions fall short. Bans shift the problem to the time before submission; detection software produces false verdicts and a climate of distrust. Both defend a format that has just lost its power to tell us anything.
„Cheating isn't new. What's new is that the old exam no longer measures anything."
What still measures
Anything that makes the path visible: interim versions instead of a final submission, reasoned decisions instead of a result, a short defense of what was handed in, work under observation, oral follow-up questions exactly where it gets interesting. Anyone who can explain their work has understood it — with or without AI.
Interestingly, that is the standard wherever much is at stake: oral exams, colloquia, case presentations. We never abolished these formats; we just considered them too costly to use at scale.
The honest price
Formats like these cost time — teachers' time, which is scarce as it is. This is exactly where AI can help: preparation, creating materials, first rounds of feedback, administration. The time gained then belongs in the conversation no machine should be having.
In the end, the question is not how to make exams AI-proof. It is which abilities count in a world with AI — and whether we ever really tested those.
Questions about this
What should exams look like in the AI era?
They should assess the path, not just the result: interim versions, reasoned decisions, an oral defense of the submission, and work under observation.
Does AI detection software help against cheating?
Hardly. It produces false verdicts and distrust without solving the real question: whether a format still measures understanding at all.
Where is the time for more demanding exam formats supposed to come from?
From relief elsewhere: preparation, materials, first feedback, and administration can be shortened with AI — the time gained belongs in the conversation.
