Two major reports warn that AI in schools may harm students' cognitive development. In Denmark, SkoleGPT is criticized for being unreliable. But the solution lies neither in bans nor in the pursuit of perfect machines. It lies in pedagogy and in human interaction.
Here is this week’s newsletter on artificial intelligence in the education system, where this time we focus on what GenAI does to our brains.
Last week saw the release of two major reports from OECD and the think tank Brookings, respectively. Both warn that if we use AI without a clear pedagogical framework, it can harm pupils’ learning. The reports conclude that we risk a form of “cognitive outsourcing”, where pupils become dependent on the technology and lose the ability for critical thinking and social empathy.
The same concerns are shared here in Denmark, where Anders Bæk warns in IT-Watch that pupils risk becoming less intelligent. At the same time, the Nordic teacher organisations have issued a joint statement making it clear that technologies must never replace the teacher.
Meanwhile, SkoleGPT was criticised by a teacher for being moralising and unreliable. Martin Exner defends the model and explains that it should be seen as a sandbox for Bildung.
The solution to all these challenges is not to ban the technology in teaching or simply leave things to chance. The answer lies in pedagogy and in the human, as both Dean Maja Horst and Associate Professor Monica Irene Krarup Olesen point out in their respective pieces. The technology of the future requires us to bring in humanistic knowledge and rethink our entire view of learning. Here they both argue that we should move away from control and towards participation.
In this week’s commentary I argue that for many years we have had the wrong view of AI, and that we now need to shift the focus from cheating to cognitive training.
Enjoy the newsletter.
The latest OECD Digital Education Outlook 2026 focuses on how generative AI can benefit teaching and learning, but only if the technology is used with a clear pedagogical purpose. The report explicitly warns that pupils who use general AI tools without a learning purpose risk becoming dependent on the technology. This can lead to reduced engagement and lack of skill development.
Conversely, the report highlights that when GenAI is used as a pedagogically integrated part of teaching, for example in collaboration-based learning environments, it can strengthen skills such as critical thinking, creativity, and argumentation. The report particularly highlights the potential of intelligent tutoring systems (ITS), which can create more engaging, adaptive, and dialogue-based learning experiences.
For educators, GenAI can increase teacher agency by especially supporting inexperienced teachers and improving their teaching effectiveness. The value is further enhanced if teachers are involved in the design and development of AI tools.
OECD recommends that GenAI be used purposefully in teaching and not as a substitute for human relationships or cognitive effort. Governments and policymakers are encouraged to invest in research into pedagogical GenAI, create frameworks for responsible use, and ensure an infrastructure in which all pupils and teachers have equal access to well-designed solutions.
https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026_062a7394-en.html
A new report from the Brookings Institution warns that AI’s current use in schools may do more harm than good, particularly for children’s cognitive and social development. The report draws on interviews and focus groups with pupils, teachers, parents, and experts from 50 countries, as well as a review of several hundred research articles. Its conclusions are described as a premortem, as AI’s full long-term effects are not yet known.
On the positive side, the report highlights AI’s potential to support language learning and writing, especially for pupils with special needs or who are learning a second language. AI can also ease teachers’ workload by automating tasks such as translation, planning, and task management. In some cases, AI can improve access to education, for example for girls in Afghanistan or pupils with dyslexia.
However, the report stresses that AI in teaching should primarily supplement and not replace the teacher. The greatest concern is that pupils lose the ability to think independently. AI can lead to a form of cognitive dependence, where pupils increasingly hand over thinking to the technology and thereby risk a deterioration in critical thinking, creativity, and subject understanding.
AI’s impact on pupils’ social and emotional development also raises concern. In particular, chatbots are designed to affirm the user and therefore do not challenge pupils in their social formation or opinion formation. This can weaken the ability to handle disagreement and develop empathy.
Another central issue is inequality, where schools with limited resources gain access to less reliable AI tools, while wealthier schools can pay for more accurate solutions. This can exacerbate social inequality in the education system.
Read more here:
See More
Anders Bæk warns in a column on IT-Watch against what he calls cognitive outsourcing, the tendency to hand over too many mental tasks to AI assistants such as ChatGPT. He believes we risk our cognitive abilities weakening when we no longer train them actively.
He particularly points to young people and pupils as a risk group. Many teachers he has spoken to in schools and educational institutions report a noticeable decline in pupils’ ability to concentrate and understand complex topics.
He therefore urges that AI be used thoughtfully, as a supplement and not as a replacement.
On the occasion of the International Day of Education, the leaders of the Nordic teacher organisations have issued a joint statement emphasising that AI may only be used as support in teaching, never as a replacement for teachers. They point out that decisions about AI in school must be made with the involvement of the teaching profession and be based on research and documented experience.
They warn against technology decisions made without teacher involvement, and against the use of AI for the surveillance of pupils and staff. The organisations demand that teachers’ professional autonomy be respected and supported through continuing professional development (CPD) and time for skills development.
A key message is that, without national strategies and resources, AI risks increasing inequality between schools and pupils. The organisations therefore call on the Nordic governments to develop shared strategies that ensure the teaching profession has real influence.
In an opinion piece in Altinget, the Danish primary and lower secondary school (folkeskole) teacher Jonas Stenberg Lauritsen launches a sharp critique of SkoleGPT. He believes the model helps to undermine core values such as intellectual freedom and critical thinking, because it tries to avoid offending rather than to enlighten. Using concrete examples from teaching, he shows how the model often responds with over-caution, factual errors, and moralising. This is particularly evident when the model is confronted with politically, culturally, or identity-politically sensitive topics. For example, it answers freely on questions about Buddhism but avoids equivalent questions about Islam.
Here he finds that SkoleGPT is not academically reliable, as in certain cases it distorts facts or gives unclear and misleading answers, especially when it takes normative considerations. He emphasises that this weakens pupils’ opportunity to learn about AI, source criticism, and bias, even though that is precisely one of the model’s stated aims.
Martin Exner, the project lead for SkoleGPT, responds to the criticism raised in Altinget by folkeskole teacher Jonas Stenberg Lauritsen about the model’s academic reliability and moralising tone. He acknowledges the criticism and stresses that SkoleGPT was not developed as a reference work but as a tool for personal formation and critical thinking. He explains that the model functions as a technological sandbox in which pupils can learn about both the possibilities and limitations of AI, including bias and factual inaccuracies.
According to Martin Exner, SkoleGPT has been developed with a focus on safety, transparency and data ethics. It is hosted in Denmark, does not log data and is open source. These are all deliberate choices to protect pupils. He acknowledges, however, that this has resulted in a model that cannot match the big AIs in complexity and accuracy.
Dean at Aarhus Universitet (Aarhus University) Maja Horst warns in an opinion piece in Altinget of the danger of repeating past mistakes, where technologies such as social media had far‑reaching consequences before society understood their full significance. If we want to avoid new technological dead ends, for example with AI, policymakers need to bring humanities and social science knowledge in early in the process, not only technical expertise.
Technological preparedness is not only about code and investment, but also about people, culture and trust. AI models trained on American texts and norms risk, for example, shaping Danish communication and values. Maja Horst also points out that AI may create new digital inequality, where only the best resourced gain access to the best solutions.
In Gymnasieskolen (the Danish upper secondary teachers’ magazine), Monica Irene Krarup Olesen, senior lecturer at Professionshøjskolen UCN (University College of Northern Denmark), writes that the debate about AI in the gymnasium has so far primarily focused on preventing cheating. This approach risks overlooking deeper questions of didactics and views of learning.
AI is not only a technological tool but also a cultural instrument that affects the way students think, work and participate academically.
When students use AI to generate analyses and arguments, it becomes difficult to sustain the notion of learning as an individual achievement, evidenced in a concrete product.
Therefore, AI in the gymnasium should not first and foremost be seen as a control problem but as an opportunity to rethink didactics and subject‑specific practice.
Monica Irene Krarup Olesen
Ever since ChatGPT was launched in November 2022, we have spent untold amounts of energy debating whether pupils are cheating. It also took up a good deal of space in the many early talks that Claus and I gave at gymnasiums around the country. In the end we decided it was a deadening discussion that derailed the whole conversation about using AI in teaching. So we turned it around, keeping the pedagogical possibilities front and centre.
Cheating is a real challenge, yes. But as this week’s reports from the OECD and Brookings indicate, we have been focusing on the wrong thing. The real danger is not whether pupils get the top grade (12 on the Danish scale) without deserving it. The danger is cognitive outsourcing: that we slowly accustom both ourselves and our pupils to letting language models do the thinking.
The reports warn that if we unleash the technology without a clear pedagogical framework, pupils will not become wiser – they will become dependent.
We see this in the debate about SkoleGPT, where people go looking for a perfect, error‑free machine – a kind of wiki that can give pupils the right answers without offending anyone. As Martin Exner rightly points out, learning happens precisely in technology’s errors and limitations. If the machine were perfect, we would not need to think for ourselves. It is the perfect machine that makes us dumber.
Professor Emily M. Bender points to a paradox: the more reliable an AI appears to be, the more dangerous it is for our thinking.
Furthermore, a system that is right 95% of the time is arguably more dangerous than one that is right 50% of the time. People will be more likely to trust the output, and likely less able to fact check the 5%.
The point is that the better AI becomes, the less inclined we are to think for ourselves. And that is precisely why we cannot leave formation to a language model. Turning cognitive outsourcing into cognitive training requires professionally skilled, upskilled teachers and clear didactics.
