Fable 5 gave me new ideas about my own field. Next time it may be ahead of me.

When a new frontier model is released, it often takes me a few days before I realize that I should run it against my standard benchmark question:

How will AI advancements affect the Swedish schools in 5–10 years? Reason from first principles.

I use this question, or variants of it, since it is an area where I have done a lot of thinking, and I can pretty easily assess the quality of the answers.

The introduction of reasoning models brought the first time I felt that an AI actually contributed something. Before that, the AIs regurgitated the general talk on social media and missed many important points. But o1, as the first example, gave some new input that was actually meaningful.

This evening I gave Fable 5 my standard benchmark question, and then spent some time discussing with it.

Fable picked starting points that were a bit surprising to me. When discussing or analyzing deep questions about education, I start with the purpose of education. Fable picked the functions that schools provide, which is a much more pragmatic choice. (The functions were knowledge building, sorting/certifying, care/socialization, and being an equalizing force in society.) Using the functions that schools serve, rather than the purpose of schools, makes a much more direct path to analyzing how AI impacts actual school practices and what mechanisms can be used to affect them.

Fable then went on to analyze how AI advancements affect these functions, including how certain or uncertain it deemed the predictions. Finally it added some extra comments on how slow schools change in comparison to AI, and how changes in the labour market may erode the primary motivation for many students. And of course a summary at the end.

What Fable answered on my first prompt, and our continued discussion, will change parts of how I describe and think about AI’s effect on education. It helped me see paths to answer or explore some of the core questions, such as how behavioural or structural changes actually may come about, and how the purpose of schools may change.

To be clear: What Fable delivered on how AI affects education exceeds almost all conference talks, panel discussions, podcast interviews, blog posts and articles I’ve consumed.

I initiated and led the team for AI and education at Sweden’s National Agency for Education, and since December 2022 I’ve invested a lot of time analyzing and thinking about how AI advancements affect education. I’ve written three books on AI and education.

I feel that I have an edge over Fable 5 when it comes to understanding these questions. I’m slightly better at identifying the core issues and the most important challenges. But I also know that I, like humans in general, am biased in this assessment. Fable gave me genuinely new angles for viewing systemic questions concerning AI and education. And I am preparing myself for that when the next generation of AI comes, it will be me that contributes to the conclusions of the AI, not the other way around.

This does not mean that AI is about to be on par with world-leading experts in all fields. I think I can rightly say that I’m one of the people in Sweden who knows most about, and has thought most about how AI affects education. But I am not a world-leading expert, and the field of AI + education is extremely nascent compared to many others.

But I do think that unless you’re considered an expert in a particular (cognitive) field, it is highly likely that the best AI is better than you in that field.

The most important thing to understand about this, is that we’re on a curve. This is the state of AI in June 2026. The same technology that in December 2022 could write half-decent school essays, and in 2019 could barely count to ten.

Understanding what this technology does to society and the world – and to education – is a great challenge. But trying to answer that is important if we are to have a fair chance at steering it in the right direction.