One day, one generative AI tool

Focus

The question before the tool

Focus of the letter 52

With the applications of the week, an instruction or a document is enough, and the result is neat. The results are useful, often impressive, and respond to a certain idea of what their users expect. But what do those who use them really expect, starting with the youngest, potential users of today and tomorrow? An answer comes to us from Geneva, in this week’s readings.

The canton of Geneva surveyed 12,400 pupils, from the equivalent of CM2 to the final year of secondary school, with questions developed with the Youth Council, an assembly of young people aged 14 to 21 set up by the Council of State. “Questioning young people by young people,” summed up Anne Hiltpold, State Councillor in charge of public education, at the opening of the round table to report on this survey (Léman Bleu, dans les lectures).

What these children and teenagers say they expect is precisely something other than a finished production. In the orientation cycle (middle school), about 70% consider it important to learn by themselves even when the AI gives the answer, and almost half are afraid of thinking less. Across all levels, only about 4 out of 10 say they know what they are allowed to do with AI. During the round table, a secondary school student described these ready-made answers that allow you to turn in a clean work and get a “nice 6” (in Switzerland, the best mark on a scale of 6), and then asked how you learn with it. His teachers, he added, sometimes advise him to use AI without always telling him how.

One of the interests of this consultation lies in its method: young people are not only the recipients of decisions, they participate in the formulation of questions. This is still far too rare. In the episode of “Grandir” of “Vers le Haut”, relayed in readings, Erwan Paitel, notably rapporteur in 2024 of the commission “AI: our ambition for France”, regrets that reports and debates almost never engage the voice of the child.

In this same episode, Cécile Cathelin, a high school literature teacher, tells us that she organizes her year like a play. In the first act, the first term, AI remains outside the classroom: we talk about it, and she invites her students to say everything, their questions as well as their fears that AI will replace them in ten years or that it will take precedence over their thinking. Only in the second act, AI enters the scene, as a “secondary character”: a companion who helps at times. Where the week’s applications immediately deliver a production, this approach takes the time to listen before bringing in the tool. We listen too little to users, let alone teenagers, and not only on the subject of AI: the news has unfortunately shown us this for several weeks…

But asking is not enough: what you say you expect and what you choose do not always coincide. On compar:IA, Internet users vote to decide between generative AI models. In a study presented by Simon Zilinskas in the readings, the compar:IA team and two researchers from Paris Dauphine show that for comparable length and presentation, a model that puts more text in bold has an 11% higher chance of winning the vote. Correlation, not proven causality, they specify. A vote also records what pleases at first glance, and this is what these AI tools take care of: a ready-to-use visual, a video, a dashboard. The Geneva student said it in his own way: a well-presented work is worth a “nice 6”, it does not necessarily show what you have learned.

The canton of Geneva does not want to stop at listening. Anne Hiltpold has announced classroom workshops, calls for projects, and then work with young people on governance and disinformation, to bring their proposals to the heads of state meeting in Geneva in June 2027, on the occasion of the world AI summit. It’s a good basis, but announcements are not yet results. Writing a clear rule for all students in the canton, training teachers, arbitrating between conflicting expectations will take time. What counts above all is the starting point: in Geneva, we started by asking.

And all of us, who test and use these applications? This question goes beyond school and affects our daily lives. After more than 1,300 tools tested, I am often asked for my opinion. The number of articles on the site does not make my opinion anything more than an opinion: that of a single person, perhaps with a sharper eye from habit, who tries a tool for a few days.

When I dismiss an application, it’s sometimes for reasons of value, and often for the user experience (do we understand what the tool does, where to start, what to do when we make a mistake?) or for the interface itself (buttons, texts, readability, screen organization, etc.). These two dimensions are verified by watching people use the tool or model, not from its designers’ computers. These rejections remain subjective, but the same reasons often come up: from what I see, too few applications seem to be tested with their target audiences before their launch. Supporting designers, from the idea to the meeting with their users, would probably help more than additional opinions.

In Geneva, it took a questionnaire and young people to develop it, and it will take time to transform their answers into decisions. In an application, an instruction is enough to obtain production. However, these technologies affect all levels of our society: who decides and who takes the time to ask those who will use them what they expect from them?

Echo: