Marc Cleiren

25 augustus 2025

At the start of another academic year, our wrestling with Artificial Intelligence continues. Use of AI has recently shown itself as the great destroyer of critical thinking—the corrupter of young minds, eroder of memory capacity. It threatens to be the intelligence-killer of an entire generation. There is truth in that. Almost all recent empirical studies, including this year’s MIT report (Kosmyna et al. 2025), and a Wharton paper (Melumad et al, 2025) confirm what many teachers have witnessed firsthand these years. 

Kenan Malik’s recent analysis on AI in The Observer (August 2025)—puts forward an uncomfortable question: “If knowledge is merely a commodity, why not cheat to bypass the need to think?” and has a warning: A culture that views knowledge as a means to an end invites the misuse of new technology. He is right. The misuse of AI is not primarily about the technology itself—it is about what education has become. Knowledge has been increasingly framed as a ‘commodity’: an instrument for more profitable economic and commercial gains, its funding wielded by non-expert politicians. Against that backdrop, the temptation to outsource our thinking to machines is not surprising.

But does this mean we (academic) teachers should give up? Should we throw our hands in the air and admit defeat? Please, no. Instead of despairing, AI can finally enable us to step up our game. Decades of carefully researched didactical knowledge are waiting to be put into practice. AI, if guided wisely, can help us finally do just that for which we hitherto had no time or other resources.

How we broke education

Since the 1970s, research by Edward Deci and Richard Ryan has shown that frequent summative testing: judging the knowledge of students and grading it, is a major way to kill intrinsic motivation. It replaces the joy of learning with a feeling that it is a ‘chore’. Still, this is what has been instilled by consecutive governments:  measure, index, optimize output, maximize profit and efficiency as the main purpose of ‘learning’. No wonder that students over the years started to call academic education ‘going to school’ defined their studies as mainly ‘homework & exams’, becoming a duty often separate from the rest of their personal life.

Let’s face it. For decades, the old fashioned frequent-grading and selection systems in schools and academic institutions have been used as ways to chase students into learning. As I see it, they are a cheap replacement for face-to-face individual motivation in teaching. This has gradually been eroding student motivation for decades.

How AI can help out

How can we hope today to rekindle the true desire to learn in our students? The answer may be relatively simple: by intelligently employing AI on a teacher level. Using didactical knowledge on constructive alignment, motivation, blended learning, scaffolding and the like to create highly able AI assistants, that have limitless time on their hands.  I want to share a couple of basic thoughts on that here, that we are currently also piloting in courses across Leiden University.

1. AI as Master-Apprentice Teacher

Lesson one: only master teachers can build expert teachers. That is true also for LLM’s. GenAI models  have potential, and are quick learners, but we must teach them the profession of teaching: aided by experts from each scientific and societal domain . The best model for this isn’t new—it’s ancient: the master-apprentice relationship. Hugely time intensive, but extremely effective (Bloom, 1984). The master has all the knowledge, but gives out only what the pupil can take in. Testing the level of the apprentice, at any step of the way, is by observing the apprentice at handling the tools (like the master blacksmith watching how the hammer is wielded). First, he requires simple tasks, mainly to be able to observe basic skills, later when the pupil is ripe, he presents more complex ones. The blacksmith, because of his nearness to the apprentice, does not need ‘intermediate  masterpieces’ to assess progress. Their eyes and ears follow the process itself. Master pieces are reserved to crown the stages, as an initiation to sets of tasks where  independent workmanship of the apprentice is necessary (we call it: a job). Before that, tests are likely to trigger the self-protective part of a person, effectively shut down the learning mindset and stimulate to (re)produce instead of think.

Human teachers in the public domain have no time to stand by their side, but AI teachers can basically be there day and night. Be all eyes and ears in the depth of night, even after heavy partying, within years, probably we will also have avatars broadly available, communicating live with the student, as well as employing speech-to-speech rather than processing only written language. This can speed up learning and motivation even more. Today, as public education is facing huge cutbacks in funding, human teacher-to-student interaction even more skimp.

2. AI as a Student’s Ally

Let us put LLM’s pre-trained ability to maximize user engagement instead of commercial use it was aimed at, to good use: maximizing student engagement. Remember your own favorite teacher? (Yes, imagine him or her right now..). Probably they were really seeing you, treating you like a person. They were able to transfer their enthusiasm onto you, making you motivated to follow their lessons, their insights, stimulating your curiosity.  

Critics worry about AI’s sycophantic tendencies—its habit of always agreeing, always flattering, and the fear that people are misinformed and led astray by this. However: relationship between teacher and pupil is the carrier of the information, relation is the moderator that decides whether information is taken in and digested or not.

3. Focus on Learning, Not Just Grading

This intuitively runs counter efficiency ideas and economic value. It also scares teaching staff to postpone the proof of the pudding of hat has been learned. However personally, I have seen many very talented, very bored students (including my own children) simply because they had to perform frequent tasks, jump through exam-hoops that had no meaning or use outside the test itself. It took their teachers and parents many efforts to refrain them from giving up the subject or even studies. Not because they were not able to, but they saw no purpose in what they were doing and no relation to a goal.  AI is hugely effective as a guide, if prompted in the right way. One that has the time for individual guidance that (academic) teachers currently have not. At the process of concise grading, it may even be better than human graders.

4. Multiple Teaching Personae

Finally, good teachers know how to shift roles: encourager, critic, coach, or Socratic guide. AI needs the same flexibility. It should first get to know a bit about the student: their interests, anxieties, hangups, level of knowledge, and learning style. It should then assess actual knowledge—never assuming—before identifying gaps and building confidence that growth is possible. From there, it can scaffold learning, gently pushing the student forward, using questions rather than commands, and adapting tone and method along the way. In other words: we need to program AI not as a single fixed tutor, but as a set of teaching personae, each suited to a different stage of the learning process.

Conclusion: AI as the Teacher We Need (to Train)

AI is not inherently a threat to education. It enables us to develop brand-new student interactions that were hitherto not possible. Also in the classroom. But AI is also the mirror that reflects to us what public education has become. If we continue to treat knowledge as a commodity, AI will accelerate the collapse of critical thought. But if we as knowledgeable teaching staff use AI to embody what we already know about good teaching—constructive alignment, motivation, scaffolding, human connection (we will all discuss in a later blog) - it can restore what our system has eroded: curiosity, intrinsic motivation, and the joy of learning.

We are at a crossroads. AI in the commercial wild can be the gravedigger of education. Paired with educational knowledge it may be its saviour. The choice lies not in the technology, but in how we, as teaching institutions, choose to invest in it. And yes, our AI pilots will meet with problems and mistakes, but we can no longer afford negligence of AI’s capabilities, nor of our own pedagogical knowledge.

Relevant References

Bloom, B. S. (1984). The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, 13, 4-16.
https://dx.doi.org/10.3102/0013189X013006004

Deci, E. L. (1971). Effects of externally mediated rewards on intrinsic motivation. Journal of Personality and Social Psychology, 18(1), 105–115. https://doi.org/10.1037/h0030644

Deci, E. L., & Ryan, R. M. (1985). Intrinsic motivation and self-determination in human behavior. New York, NY: Springer Science + Business Media.

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025, June 10). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task [Preprint]. arXiv. https://arxiv.org/abs/2506.08872

Melumad, Shiri and Yun, Jin Ho, Experimental Evidence of the Effects of Large Language Models versus Web Search on Depth of Learning (2025, January 20). The Wharton School Research Paper, Available at https://dx.doi.org/10.2139/ssrn.5104064

Malik, K. (2025, August 1). AI thrives where education has been devalued. The Observer.
Retrieved from https://observer.co.uk/news/opinion-and-ideas/article/ai-education-knowledge-technology

This blog post represents only the author's insights and opinions and does not represent the official position of Leiden University. Some editorial revision suggestions for this article were provided by Julian van der Kraats. Some reference & heading suggestions were provided by ChatGPT5 on August 25, 2025; image created by a GPT-5 text interpretation of the current blog and Midjourney v7.