France says that starting in the fall of 2027, every student entering “seconde”, roughly the U.S. equivalent of 10th grade, will get artificial intelligence lessons as part of the national curriculum.
But the country’s leading computer science society is warning that the rollout, as described so far, risks turning into a glossy tour of trendy tools instead of a real education in how AI works, and why it fails. The Société Informatique de France (SIF), a professional group of researchers and educators, is urging officials to tighten the content, train teachers at scale, and avoid widening gaps between well-resourced schools and everyone else.
Scientists warn: Don’t confuse “using AI” with understanding it
Sommaire
SIF’s core complaint is simple: teaching teenagers to prompt a chatbot or generate an image isn’t the same as teaching AI. If the course leans too heavily on everyday apps, AI-assisted writing, image generation, automated summaries, students may walk away with buzzwords, not knowledge.
The group argues that even an introductory class needs a conceptual spine: data, models, training, bias, evaluation, and limits. Data, in particular, is the foundation of modern AI systems, and it comes with real-world problems, quality, representativeness, and traceability, that shape what these systems can and can’t do.
SIF also worries the new requirement could blur a basic distinction: AI is one slice of computer science, not a replacement for it. Without clear links to fundamentals like algorithms, data structures, probability, and optimization, an AI unit could become a standalone “tech culture” module, interesting, maybe, but disconnected from the skills students would need for more advanced coursework later.
That risk is amplified by uneven conditions on the ground. Some French high schools already have trained staff, updated computer labs, and established digital projects. Others struggle with basic tech access. A one-size-fits-all mandate, SIF argues, needs precise, scientifically vetted content and realistic lesson plans that work in mixed-ability classrooms.
The biggest bottleneck may be teacher training, and time
Making AI a standard part of 10th grade means training a lot of teachers quickly. And the instructors won’t necessarily be computer science specialists; “seconde” is part of a common core, and schools may assign the material across different departments.
SIF is flagging a familiar trap in tech education reforms: launch fast, fix later. AI changes quickly, and a one-off training session built around broad overviews could leave teachers unprepared for the questions students will ask, about reliability, so-called “hallucinations,” and the difference between recommendation algorithms and generative models.
There’s also the workload. Designing an AI lesson that’s accurate and age-appropriate takes more than understanding the topic; it requires turning complex ideas into classroom activities. Without strong, ready-to-use materials, preparation falls on individual teachers, an approach that can deepen inequality between schools with experienced teams and those without.
Then come the practical constraints: many online AI tools change terms of service, require accounts, or impose age limits. In a school setting, that raises issues around student privacy, data protection, oversight, and whether students can actually use the tools themselves, or just watch a teacher demo them on a projector.
Finally, SIF notes that implementation won’t rest only on teachers. Principals and IT staff will face decisions about devices, networks, and software, choices that depend on budgets that vary widely from school to school.
AI literacy is also civics: bias, automated decisions, and misinformation
SIF’s argument isn’t just about technical rigor. The group says schools have a civic responsibility to prepare students for a world where automated systems increasingly shape what people see and how they’re judged, content feeds, hiring filters, scoring systems, and decision-support tools.
That means teaching limits and uncertainty, not presenting AI as a magic black box. Bias is a central lesson: models learn patterns from training data, and those patterns can encode stereotypes or discrimination. Students should learn how bias can emerge, how to spot it, and why “just remove bias” is often harder than it sounds.
Misinformation is another urgent front. Tools that generate convincing text, images, and video make it easier to mass-produce fakes. An AI unit, SIF argues, should strengthen verification habits, how synthetic content is made, why it can sound confident while being wrong, and why sourcing still matters.
The goal, in SIF’s view, is a student who understands that AI systems optimize for patterns, generalize from examples, and can fail in systematic ways, knowledge that’s useful far beyond the classroom.
What should a 15-year-old learn first: flashy tools or fundamentals?
The curriculum challenge is real: limited class time, wide variation in student backgrounds, and a topic that can get technical fast. Hands-on demos can hook students and spark discussion. But if the course stops there, SIF says, it risks teaching “AI as spectacle” instead of AI as a set of testable ideas.
A workable compromise, many education experts suggest, is to start with concrete examples and then climb toward principles. Show an image classifier, then introduce training data, test sets, error rates, and overfitting. Use a text generator, then explain probability-driven prediction, and the gap between fluent language and factual truth.
SIF is pushing for minimum precision even at an introductory level: clear definitions, basic distinctions (like supervised vs. unsupervised learning), what a “model” is, and what it means to evaluate performance. The point isn’t to turn 10th grade into a college course, it’s to avoid explanations that mislead students, like implying the system has intentions or confusing correlation with causation.
Assessment matters, too. If AI becomes part of the official program, schools need ways to test understanding beyond vocabulary quizzes, through analyzing outputs, identifying bias in examples, or comparing performance across approaches.
With 2027 approaching, SIF’s message is that France can absolutely teach AI to every teenager, but only if the government treats it like a serious science subject: clear standards, coherent progression, credible training, and classroom-ready resources. Otherwise, the country may end up with a national mandate that produces wildly different results depending on the school a student happens to attend.
