Alexander Klöpping with Eva, a good teacher doesn't give the answer straight away
This week on Eva on NPO 1, the topic was AI in education. Alexander Klöpping and educationalist Barend Last joined the discussion, and in fourteen minutes, almost everything we at CivAI work on every day was covered. Students admitting that three-quarters of their work comes from AI, a lecturer who only checks AI-generated texts, and an exam that no longer measures what it is supposed to measure.
One sentence from Alexander stuck with me. In a single breath, he explained where things go wrong, and where the difference lies.
The highlights to rewatch
Don’t have a quarter of an hour? These moments are worth your time. The links take you straight to the right moment in the broadcast.
From 1:58, a university lecturer wonders whether he should just use AI to check AI-generated texts.
From 4:36, Barend Last on why the problem lies with the exam, not the student.
From 6:13, students in Utrecht explain how much of their work comes from AI.
From 9:00, Alexander demonstrates an AI agent that watches lectures, writes papers, and submits them itself.
From 10:58, the difference between AI that helps and a teacher who encourages you to think.
From 12:39, the dream of a personal tutor for every student.
A good teacher doesn’t give the answer straight away
Alexander showed just how far things have come. There are tools being sold that log in to your university, keep track of your assignments, write your paper, run it through an AI detector, make the text more human, and submit everything on time. And if the grade is disappointing, they email the lecturer for a resit.
Afterwards, he said something I’ve been trying to explain for months.
"AIs are fundamentally built to help you. If you ask a question, it will give you the answer. Whereas a good teacher, of course, won’t give you the answer straight away. They will complicate it, help you to think."
That’s exactly it. A chatbot that delivers takes over the thinking. A good teacher asks a question in return, gives a hint, and lets the student take the next step themselves. That’s not a technical detail; it’s the difference between learning and simply submitting work.
The problem is with the exam, not the student
Barend Last put it sharply. If an exam no longer measures what it is intended to measure, then the exam is not fit for purpose. He believes that reporting on AI-generated dissertations too often points the finger at the student, while many lecturers have been working hard for years to redesign their assessments.
He also gave an honest explanation for students’ behaviour. In a system where the grade is all that matters, taking the path of least resistance is a perfectly rational choice. His solution is assessments that are much more dialogic, closer to the student.
I completely agree. I recently wrote an extensive piece about this, following RTL’s investigation into master’s theses. Read it here, including research showing why detection tools should never be the sole evidence.
What happens between question and answer
In my book The Art of Living Together with AI I describe the HEAR framework, four layers for human-centred work with AI. It starts with the human’s intention (Human Intent) and the way you interact with the system (Expressive Interaction). It ends with someone who bears the responsibility and can explain how the decision was made (Reasoned Responsibility).
In between lies the layer Alexander was really talking about: AI Mediation. What does the system do between the question and the answer? Does it deliver immediately, or does it leave space for the student to think for themselves? That layer determines whether AI enhances learning or replaces it.
The agent Alexander demonstrated removes the human from both sides. No personal intention, no personal judgement. Good education with AI does precisely the opposite.
The dream of a personal tutor is already possible
At the end of the broadcast, Alexander outlined what AI can also be. Tutors who think along at the pace and knowledge level of each student, fully tailored. He called it a dream, but saw that in practice things are mostly going the other way, with bans and AI-free exams.
That dream need not wait. With EduGPT, we have built exactly that layer between question and answer. Our Critical Thinking Engine does not deliver ready-made work. It asks what the student has already done, what was difficult, and what the student thinks themselves. Then comes a hint, a structure or an example, but the thinking remains with the student.
A school or course decides for itself, per group or assignment, when that guiding layer is enabled. And the teacher remains the assessor. AI can give quick feedback, but the judgement remains with the teacher. All of this runs on European infrastructure.
Information is everywhere. Guidance is not.
Back to the drawing board together
Barend Last called it a wicked problem, a challenge without a simple solution. And so it is. No tool will solve it for us, not even EduGPT. It requires teachers to redesign their assessments, courses that dare to make choices, and AI that supports learning rather than taking over.
Would you like to know what to look out for when choosing AI for your education? In this article we list seven characteristics of educational AI.
I’m curious how you are dealing with this. Let me know, or visit edugpt.nl to see how it works in practice.
Source, Eva (AVROTROS, NPO 1), "How AI is changing our education", October 2026, with Alexander Klöpping and Barend Last. Quotations have been transcribed from the broadcast.
Illustrative image, created with AI