Here’s this week’s news, focusing on an exam system in crisis, while also delving into concrete solutions, Sweden’s AI reform, and a new study that questions AI’s ability to reason.
We have gathered a range of news and articles about AI in education, and this week we focus on the examination system. There have been quite a number of contributions from many different quarters, but what they have in common is that there are major challenges in continuing with an examination system that is not equipped to handle AI.
In teaching there are also challenges in how to assess homework when the student, outside the teacher’s control, can simply use AI for the tasks.
More broadly, AI is a technology that has affected the entire education sector.
We also look to the Nordic countries, where Sweden has an interesting approach to making the technology accessible to everyone – could we do the same in Denmark?
An interesting debate last week concerned whether AI can reason.
Additionally, we have this week’s news from the tech world. Finally, you will find the articles we did not have room to elaborate on.
At Viden.AI, last week we published the first part of a three-part article series examining the challenges that have arisen in the wake of ChatGPT. If you have not read it, you can find it here:
Per Størup Lauridsen
Happy reading!
Last week, the chair of Gymnasieskolernes Lærerforening (GL), Anders Frikke, criticised Børne- og Undervisningsministeriet for not doing enough to secure exams, and said that the initiatives launched are inadequate in relation to the challenges the sector faces.
He points out that ChatGPT and similar tools threaten students’ deep thinking, which otherwise arises when you write longer, coherent texts yourself. At the same time, he highlights that written homework in particular has become meaningless, because the teacher has no way of assessing whether the student wrote the assignment themselves. He is also sceptical about returning to paper and pencil, as is happening in the language subjects. (Read more in last week’s newsletter: https://viden.ai/nyhedsbreve/ugens-nyheder-mellem-ai-begejstring-og-behov-for-kritisk-forstaelse/#eksamen)
Like many others, he therefore calls for a new form of examination that takes account of the use of AI without allowing the technology to determine the content.
6 June 2025 at 11:02
In Politiken, Anders Frikke, together with Joachim Ekstrand Federspiel, chair of Danske Gymnasieelevers Sammenslutning (the Danish Upper Secondary School Students’ Association), argued that teaching itself should be divided into modules with and without AI. There is a need for clear guidelines so that both teachers and students know how AI may be used.
Like many other educational institutions, NEXT Uddannelse København is also experiencing a growing challenge with students using AI in connection with submissions. In Berlingske, Ole Heinager, director at NEXT, and Peter Bruus, Head of AI, explain how teachers find it difficult to assess students’ actual academic level.
They therefore propose moving the entire basis for awarding grades away from homework and instead using the classroom as the primary assessment setting. This means the teacher can better follow how the student works, both orally and in writing, under controlled conditions.
Ole Heinager, director, NEXT Uddannelse København, and Peter Bruus, Head of AI, NEXT Uddannelse København
After two decades as an upper secondary school teacher, Fredrik Werenberg Poulsen reflects in Gymnasieskolen on the impact of digital developments on teaching and highlights the need to update the education system in light of AI. He points out how the arrival of technology, from PCs and tablets to smartphones and AI, has gradually changed the classroom reality, often without the education system keeping up. This has led students to log out of the physical space and into digital worlds, challenging both learning and presence.
Read the full piece in Gymnasieskolen:
Fredrik Werenberg Poulsen
On the blog AI i skolen, Mikkel Aslak has updated his popular overview poster of AI tools for teaching.
The new classroom edition focuses exclusively on approved tools for students, while the updated teacher poster includes new tools from TA Padlet, which are considered particularly interesting.
All three posters are available in full resolution and can be edited in Canva, allowing for local adaptation.
Mikkel Aslak
Ove Christensen has written a review of the book 40 AI-aktiviteter til din undervisning: Generativ kunstig intelligens i efter- og videregående uddannelse.
Read his review of the book here:
Sweden has launched an ambitious national AI reform that will provide free access to advanced AI tools for 2.3 million citizens, including public employees, teachers, students and NGOs. The initiative was presented by Prime Minister Ulf Kristersson at the Sana AI Summit in Stockholm and is inspired by the 1990s home PC reform, which played a major role in the digitalisation of Swedish households.
The aim of the new AI reform is to democratise access to AI and strengthen Sweden’s global competitiveness.
Through a collaboration with Sana and the foundation Svenska AI-reformen, users gain access to Sana Agents, a platform that uses leading language models from, among others, OpenAI, Google and Anthropic.
Irina Engeness, professor and researcher in digital learning, warns against meeting AI with fear and control in school.
In an article, she points out that AI is already affecting teaching and will, in future, change both the teacher’s role and our understanding of learning. The technology can both free up resources and support students with special needs, but it can also undermine critical thinking and deep subject learning if used unreflectively.
She urges that the education sector should have a central place in Norway’s AI strategy.
Irina Engeness
The author Salman Rushdie recently expressed concern that AI might one day write books with the same quality as human authors, while at the same time rejecting the idea that AI has a sense of humour. However, a satirical remark from ChatGPT suggests the technology may be closer than many think. Language models such as ChatGPT are built for pattern recognition, and it is therefore not unlikely that, in time, they will be able to reproduce the linguistic patterns that evoke laughter, sorrow or reflection in readers.
Marta Sørensen
A new study from Apple shows that even the most advanced AI reasoning models, such as Claude 3.7, DeepSeek R1 and Google’s Gemini Thinking, break down when faced with complex logic puzzles. This suggests that the models lack perseverance and strategic flexibility, traits normally associated with genuine reasoning.
The study raises questions about how far we really are from artificial general intelligence (AGI), and whether current models in practice merely mimic patterns of thought without understanding them.
AI researchers such as Gary Marcus point out, however, that this does not necessarily mean that the models cannot think, but rather that, like many people, they have clear limits to how much they can cope with. For example, many people cannot solve the Tower of Hanoi with eight disks without mistakes either.
Cecily Mauran
Here is a brief overview of new or improved AI technologies that emerged last week.
Casper Ravnsted-Larsen
Jess Weatherbed
Matthias Bastian
Rising
