Focus of the letter 36

While testing Honen this week, I asked for the generation of a course on generative AI in education as I am used to doing on this type of application. The tool generated in a few seconds a complete sequence: text to read, flashcards, quizzes, podcast, comic book. Complete, structured, progressive and presentable. I hardly intervened in the didactic or pedagogical choices, nothing encouraged me to do so. This is not a reproach to the tool, which is rather well thought out by the way, it is an observation on what happened, or rather on what did not happen.

Knowly does something similar: we submit a topic or sources to it, it organizes a path, generates pages linked together, suggests a next step after each generation. It’s a wiki that is built without the need to build it.

Replit Slides and Presenton, on the other hand, generate presentations from an instruction: the plan, the slides, the images, the formatting. You choose a design, you validate a plan, and the slides arrive one by one. The result is clean and shareable in one click.

These four tools do not do the same thing, but they share one thing in common: they produce the deliverable without any action on the process.

Holotab, in letter 33, asked the question of the level of autonomy that we decide to give it. Here I think that the question has shifted: we don’t really choose anymore. We let it happen because it’s there, because it’s fluid, because nothing encourages us to do otherwise, and let’s say it, because it’s effective and very practical…

And it is in this process that something happens. Building a course is first of all asking ourselves what our needs are, where to start, what is important and what is not, what we need to understand before we can explain and make people understand. To make a presentation is to find out how one idea leads to another, how to make visible a progression that may not have been quite clear in one’s own head yet. This work of preparation and shaping is a constituent part of reflection and understanding. We only really understand something by trying to convey it and when we are able to explain it. The friction of the process is ultimately a bit like an imprint that is imprinted little by little.

In the readings of Letter 35 , I shared with you a meta-analysis of more than 800 academic articles on AI in primary and secondary education conducted by the SCALE project at Stanford. One of the conclusions points out that the gains obtained with the help of AI often disappear when the tool is no longer there. This is an observation of what the tool does to learning when it takes over the process: the crutch works but when you take it off, you don’t work any better than before.

The Obvia 2026 report, which you will find the analysis of L’école branchée and the link to in the readings, also states it: succeeding in a task with AI does not automatically lead to learning. There is the task accomplished, and there is what we retain from it. The result of production and the process that leads to it are not the same thing, and the tools do not distinguish between them.

The Ipsos poll for THEIA takes up these threads implicitly. 63% of young people aged 18 to 25 believe that exams in France do not reflect their real level, but at the same time, 66% think that you need to learn even more than before to stand out from AI. And only 36% think that it is less important to train with an AI that is capable of performing tasks. This is not an abandonment of effort, it seems to me that it is the revelation that something no longer fits between what is required to produce and what we really need to understand, and perhaps a search for what effort must now produce in a world where the tool can produce the result.

The tools are effective. The results they produce are often better than what we would have done on our own and do it faster. So the temptation to let them do it is perfectly rational, almost justified? And maybe that’s exactly why the question deserves to be asked clearly: when Honen generates my course and Presenton generates my presentation, I have a course to transmit. But have I learned anything more than before I started?