Ai goes into the classroom and makes maths grades soar. But without algorithm learning collapses
In the OECD's Digital Education Outlook 2026 study, however, artificial intelligence if used well has enormous potential to help both teachers and students
In an experiment cited in the OECD's latest Digital Education Outlook, access to generative artificial intelligence tools boosted students' maths performance by around 48%. But when the Ai is taken away, like training wheels on a bicycle, performance plummets: -17%. A cognitive doping effect. It works but you don't learn. This is where we should start to understand the impact of AI on schools.
High performance, fragile learning
The message from the OECD is clear: doing better does not mean learning more. AI is great at improving performance in the short term. It solves exercises. It suggests steps. It compiles answers. Like a universal calculator of thought. The problem is that it often replaces cognitive effort instead of training it.
In the case cited by the study, students with access to GPT-4 perform better on tests. But once the digital assistant is switched off, they are left with weaker skills than before. The AI did the job. The brain didn't. It is the difference between learning to drive and being driven. The thesis of the OECD study is not destructive towards this technology. On the contrary. When AI is designed as an educational tutor, and not as a shortcut to the right answer, the results change. In the same experiment, an 'educational' version of AI - designed to guide the student step by step - improved performance by 127%. And without such marked boomerang effects. Translation: it matters how you use the algorithm, not just whether you use it.
AI that asks questions, that prompts reasoning, that slows down instead of speeding up, can reinforce deep learning. AI that gives ready answers is a crutch. Useful. But dangerous if you never take it away.
Teachers increased, not replaced
On the teacher front, the numbers are just as interesting. In the UK, the use of AI tools reduced the time spent by science teachers preparing lessons and materials by 31%. In other studies cited by the OECD, inexperienced tutors supported by AI increased exam pass rates by 9 percentage points.
