Artificial Intelligence and Educational Institution

The announcement in September 2025 by the Ministry of National Education of an artificial intelligence intended for teachers constitutes a good first step, which is necessary. For my part, I suggest going further, through the creation of a language model for students and for external use. There is in my view an issue of transmitting republican values and preserving the intellectual autonomy of citizens, here is why and how.

An encouraging but incomplete first step

The ministerial announcement of a “sovereign” artificial intelligence intended for teachers deserves to be welcomed. This tool, planned to support teachers in preparing their lessons and analyzing learning, testifies to a beginning of institutional awareness facing today’s challenges, just as the framework for use “AI in Education” published by the Ministry of National Education in June 2025. The investment of 20 million euros in AI and the stated desire to create a “sovereign, open and evolving” tool undoubtedly constitute an important step.

However, this initiative, as commendable as it may be, reveals through its very limitations the scope of the challenge that awaits us. By focusing exclusively on teachers, it neglects what I consider an essential reality: the students! They are already massively using generative artificial intelligences in their educational journey. ChatGPT, Claude, Gemini and others have become the invisible learning companions of an entire generation, shaping their representations of the world, and this without any institutional or democratic control, because these language models belong to private companies, most of them American.

This asymmetry poses a fundamental problem. On one side, we train teachers to use tools that comply with republican requirements; on the other, we leave students to themselves facing the opaque logic of multinational companies whose economic and ideological objectives have no reason to coincide with the French educational project. It’s as if we equipped professors with certified textbooks while letting students draw their knowledge from an anarchic and uncontrolled library.

The new educational paradigm: beyond pedagogical assistance

In my article Education in the Era of Artificial Intelligence, I proposed pedagogical approaches to integrate new technological realities into educational approaches. Here, I wish to contribute to the institutional dimension, the State’s responsibility, not only to equip its teachers, but above all to guarantee students access to artificial intelligence tools that comply with republican values.

The French school institution traditionally structures its action around national programs that find their embodiment in textbooks. These, although published by private actors, remain strictly framed by the requirements of the national educational project. As Condorcet already reminded us in his Five Memoirs on Public Instruction (1791), the republican school has the mission of “training enlightened citizens”, capable of exercising their critical judgment autonomously. This mission becomes impossible if we do not contribute to the cognitive tools that our students use daily. The minister states that AI is “an auxiliary brain” that must not “dispense with making your brain work”. But how can we guarantee this critical vigilance if we do not offer students tools, which they can use autonomously, that comply with our pedagogical requirements? Because if we don’t do it, they use market tools, which are in no way adapted to our pedagogical objectives. It’s not about normalizing, but about being present with our values in the current cognitive space.

The illusion of training: why educating about biases is not enough

The ministerial will affirmed in the framework for use “AI in Education” to “educate students about AI biases and limitations” constitutes a necessary but dramatically insufficient approach. It’s like claiming to teach critical thinking about media while leaving students to get their information exclusively on unregulated social networks. Critical training can only be effective if it is accompanied by concrete and accessible alternatives.

Current large language models, whether “free” or proprietary, are the product of companies whose economic logic inevitably shapes the content produced. As Cathy O’Neil analyzed in Weapons of Math Destruction (2016), these algorithms are never neutral: they encode the values and biases of their designers. Their apparent free nature masks economic models based on data exploitation and the subtle orientation of cognitive representations.

Also, these tools can create cognitive dependence that progressively transforms the way young people build their relationship with knowledge. They no longer seek to understand but to obtain immediate answers, no longer develop their synthesis capacity but externalize this function to the machine. This evolution, if not institutionally accompanied, risks creating a generation that is intellectually dependent and vulnerable to algorithmic manipulation. Even if other modes of reasoning that escape us are also being created, as I develop in the article Little Thumbelina 2.0.

Towards a comprehensive public service for educational artificial intelligence

My proposal therefore goes beyond the current ministerial initiative: the Ministry of National Education must in my view build not one but two complementary large language models. The first, already announced, for teachers; the second, equally crucial, for students. And indeed, the two could perfectly well be combined. This latter would constitute the contemporary equivalent of the textbook: a tool that guarantees students access to validated knowledge, structured according to the requirements of our programs, and carrying republican ethics.

This educational artificial intelligence for students would not aim to prohibit the use of other tools, which would be illusory and counterproductive. It would rather involve offering a public, free and pedagogically designed alternative. A tool that would not content itself with providing ready-made answers but would guide the student in constructing their reasoning, invite them to question their sources, develop their critical thinking while transmitting fundamental knowledge to them.

The technical architecture already exists in other fields. The example of the National Audiovisual Institute, with its highly successful online release of French television archives from 2007, demonstrates the State’s capacity to create large-scale digital infrastructures in service of the common good. Similarly, public digital libraries like Gallica have proven that it was possible to reconcile technological ambition and public service mission.

The democratic stakes of cognitive sovereignty

The creation of such an ecosystem of educational artificial intelligences raises issues that go far beyond the school framework. It’s about preserving what I would call our “cognitive sovereignty” against global technological giants. As Hannah Arendt emphasized in The Crisis in Culture (1961), education has the mission to “preserve the world against the mortality of its creators and inhabitants” by transmitting a cultural heritage to new generations.

In our context, this means not abandoning this transmission to commercial logic and the cultural biases of large digital platforms. A public language model would guarantee that the cultural references, historical examples, moral values transmitted to students correspond well to our societal project. It’s not about imposing a single way of thinking, students remain free to use other tools anyway, but about offering a common reference, a shared foundation in the contemporary informational ocean. It’s about a cognitive editorialization project.

This approach is part of a broader vision of what the public digital infrastructure of the 21^st century should be. Just as the State guarantees access to drinking water or electricity, it should guarantee access to reliable and ethical cognitive tools. This is a matter of social justice as much as national sovereignty: all students, whatever their social origin, should have access to the same tools of excellence to build their intellectual journey.

A realistic ambition for a democratic emergency

Some will object that this proposal is too ambitious, too expensive, too complex to implement. However, the 20 million euros announced for AI intended for teachers prove that resources can be mobilized when political will exists. The development of a model for students, building on the infrastructure already created, would represent a marginal investment given the stakes.

Moreover, several countries have already engaged similar reflections. Finland, with its national AI education program, South Korea with its massive investments in educational technologies, show that this ambition is not only realistic but necessary to maintain educational competitiveness in the contemporary world. France, with its tradition of mathematical and computer excellence, actually has all the assets to become a pioneer in this field.

The technical complexity argument doesn’t hold either. The skills exist in France, in our research laboratories, our grandes écoles, our technology companies. What’s missing is the political vision and the will to mobilize these resources in service of an ambitious educational project. As Victor Hugo wrote, “nothing is more powerful than an idea whose time has come” (1877). The time has come to provide our educational system with 21^st-century tools.

Beyond the tool, a civilization project

The current ministerial initiative constitutes a first step that must therefore be welcomed. But let’s remain clear-sighted: equipping teachers without offering students adapted tools amounts to digging a generational and cognitive gap that can contribute to weakening social cohesion. The world has profoundly changed. Today’s students grow up in an environment where artificial intelligence is omnipresent. Our responsibility is not to deny this reality or to submit to it, but to transform it into an educational opportunity.

The educational artificial intelligence that I call for would not be a tool of control or uniformization, but an instrument of intellectual emancipation. It would offer each student the possibility to explore knowledge at their own pace, according to their specific needs, while guaranteeing the quality and relevance of the proposed content. It would embody what John Dewey called in Democracy and Education (1916) a “progressive education”, capable of adapting to individual needs while pursuing collective purposes. Furthermore, the use of this tool could extend beyond students, into general civic life. The Ministry of National Education would then once again have influence throughout society.

Beyond technical and financial aspects, it’s our very conception of education that is at stake. Do we want a generation trained by opaque algorithms designed in Silicon Valley or Shenzhen? Or do we prefer to take control of our educational destiny by creating the tools that will allow us to transmit our values, our culture, our vision of the world to future generations? The answer to this question will largely determine what France will be tomorrow.

The current absence of such a global project constitutes in my view a real flaw in our educational system. It’s time to move from AI for teachers to AI for all education stakeholders. This is our best chance to preserve the republican ideal of emancipatory education in a world dominated by private algorithms. History will judge our generation harshly if we leave our children intellectually disarmed facing the cognitive challenges of the 21^st century. There is still time to act, but this window of opportunity will not remain open indefinitely.