Artificial intelligence (AI), and especially generative artificial intelligence, its most visible branch, is already widely used by students – from middle school to higher education in increasing proportions – and by teachers.- Will AI help schools overcome their crises, or will it prevent young people from thinking for themselves and acquiring real skills, turning teachers into mere tutors? Can it help reduce inequality? Do we still have a choice, or must we resign ourselves to the inevitable colonization of education by AI?
- Faced with these questions which can be distressing for young people, their families and educators, there is a rather reassuring proposal: that of “bilingualism”, speaking AI as one speaks a foreign language.
The stance to adopt in education and training regarding artificial intelligence (AI), particularly generative AI, is the subject of lively debate in the educational world, both in France and abroad . These debates are all the more complex because they involve different, even divergent, conceptions of AI, and because the technology is constantly evolving, producing increasingly powerful artificial intelligence systems. Furthermore, these systems are integrated at various levels of the educational process , from the classroom to the transmission of grades, from the creation of educational materials to the administration and evaluation of institutions and the national education system.
The central question, however, is quite clear: is the use of artificial intelligence systems, particularly generative intelligence (ChatGPT, Claude, Perplexity, Gemini, Mistral Vibe…), but not exclusively, a good thing for learners, a bad thing, or both? And the same question applies to teachers. There is no single answer to this question. On the other hand, there is a simple way, at least in principle, to address the situation: what I call “educational bilingualism.”
What are the benefits for learners?
Let’s start with the learners and the benefits that an artificial intelligence system can bring. It identifies resources (texts, images, videos, documents of all kinds) relevant to the task at hand; it explains and, if necessary, translates these documents; it helps in the creation of texts, images, plans, presentations or takes care of it entirely; it answers questions and competently supports an in-depth conversation; it offers personalized monitoring or a study plan and creates the corresponding course materials and exercises.
Furthermore, it provides learners with disabilities, as well as those whose language and culture of origin differ from those of the institution, with tools enabling them to participate in the educational process.
In general, it can thus contribute to reducing inequalities of various origins.
Of course, there is no guarantee that existing artificial intelligence systems will perform these functions satisfactorily, and their weaknesses are rightly highlighted. But it is not clear that these weaknesses negate their current benefits, nor is it absurd to think that they can be overcome.
The dangers inherent in the very principle of AI intervention in education
Of a completely different gravity are the dangers inherent in the very principle of AI intervention in education (apart from the psychological and sociological effects of using AI which go beyond its educational functions and raise other questions).
A preliminary observation is that it is not enough to examine the immediate consequences of the use of artificial intelligence systems: as with any powerful technology, its effects must be evaluated over the long term .
There is, first and foremost, a legitimate concern about the risk of learners losing their skills : by entrusting AI, day after day, with the execution of cognitive tasks inherent to the educational process, they do not learn to perform these tasks themselves. Initially, they still understand what these tasks consist of. But soon, they disappear from their world: what school or university provides—namely, the ability to think in its various disciplinary forms—is replaced by the ability to type the correct command on a keyboard. The idea of personal effort has vanished.
A second danger is the weakening of critical thinking , caused both by automation (the frequent tendency to trust a machine’s decision more than one’s own judgment and other sources of information) and by the lack of intellectual resources to exercise it.
A third risk is reducing, in the minds of learners, the teacher to the role of assistant to AI: he ceases to be a model, the guarantor of the seriousness of the educational undertaking and the certified source of the knowledge process.
A fourth problem is that of cheating or more precisely plagiarism: what the learner provides is not the result of his own thinking, but the reproduction, possibly disguised, of that of the artificial intelligence system consulted.
Finally, the social dimension of education can disappear when AI fragments it into as many individual processes as there are students: the classroom is replaced by a battery of private lessons, which can only increase inequalities.
From the teachers’ perspective
For teachers, the situation is different, but it presents the same duality.
The teacher could lose the ability, for example, to design a lesson; then—as a second step—to no longer even imagine that a lesson can or should be designed, since it would suffice to ask AI to provide one. Their critical thinking skills could weaken. They would become accustomed to relying on AI, thus abandoning their truly authoritative role, not as the holder of absolute knowledge, but as the guarantor of the pedagogical process and, consequently, their social role. They would be in a position of weakness in the eyes of parents and the administration.
And some teachers express fear, mistrust or repugnance towards artificial intelligence systems or AI in its very design.
So here we are, at a crossroads: in the interest of learners or in that of teachers – which do not necessarily coincide – should we resort to AI despite the risks and dangers it presents, or should we renounce the benefits it brings, because of these risks and dangers, and perhaps also because of the uncertainty we have about the real effects of artificial intelligence systems, in the short and long term, and their evolution in the years to come?
Letting the teacher decide: a majority answer, but one that has limitations
The official response is to delegate the decision to each teacher in the exercise of their duties. Occupying the sole center of the educational system, they decide, on a case-by-case basis, whether or not to use AI, to deploy a specific artificial intelligence system for their entire class or course, and finally, whether or not to allow students to rely on a general artificial intelligence system for their independent work. In any case, they remain “in the loop,” in a position of control or at least regular oversight.
This seemingly sensible and universally accepted response has its limitations. It leaves the problem unresolved, leaving each teacher to solve it in their own way. On what basis are they supposed to determine their policy? Do they have the necessary skills to make an informed choice? How can we avoid the fragmentation of tools, methods, and practices, which can only exacerbate inequalities?
But the main flaw in this majority position lies elsewhere: by recommending the incorporation of AI, in varying doses, into the traditional teaching process, it weakens it by exposing it to the aforementioned risks, while simultaneously preventing AI from fully realizing its potential. Instead of working in synergy, the two approaches hinder each other, and the contribution of each, positive or negative, is impossible to measure except intuitively.
Towards “digital bilingualism”: learning to do with it and learning to do without it
On the contrary, I propose to institute a “digital bilingualism” : to establish, at all levels from middle school onwards and in all disciplines, courses in which digital tools, including AI, are banned, and in parallel courses where they are deployed as widely as possible.
These two-track courses, “without” and “with”, would be designed on the model of bilingual institutions, in which certain subjects or parts of the course are taught in one language and others in another, it being understood that the learners themselves use the corresponding language in their interventions and deliverables.
This digital bilingualism would offer decisive advantages of three kinds.
First, “digital-free” teaching would provide the cognitive and ethical foundations of the traditional approach, which has proven its worth for centuries: basic knowledge would be acquired and, moreover, the standard of truth—that is, the value of pursuing disciplinary knowledge for its own sake—would be instilled by example. “Digital-enabled” teaching would give learners and teachers the opportunity to develop solid digital skills that will be necessary in their professional lives, as well as in their civic and private activities. And the transition from one modality to the other would give both teachers and learners the opportunity to progress independently in understanding the advantages and limitations of technological tools and to cultivate the critical thinking that is rightly considered so important.
Secondly, the participants, in either role, would be protected from three risks:
- that of the unavailability of digital tools, following a technical failure, an economic crisis, a natural disaster, a war: they would have learned to do without;
- that of the dehumanization that could result from a practice entirely dependent on digital tools; in particular, the pleasure of the profession (for teachers) and that of participation in the social life of the class (for learners) would be preserved;
- that of irreversibility: ignorant as we are of the long-term effects of the use of digital technology on all learners or on certain populations , we might want to ban it, or at least limit it drastically; we would be able to do so, having retained the ability to do without it.
Thirdly, we would be better able than in a mixed system to assess the contributions and disadvantages of technologies, having given them every chance in the “with” sectors, and being able to compare them to the “without” sectors.
AI in education: good or bad thing?
The fact that the majority of teachers, according to recent surveys, are willing to use AI in their practice, provided they can do so as they see fit, and that many find definite advantages in it, does not preclude others from being strongly hostile to AI, nor does it prevent some of those who are not currently hostile from becoming so, as AI is beginning to get a bad rap in public opinion, which is very divided . Bilingualism would allow everyone to practice their profession in accordance with their convictions: those who are totally opposed to AI would only teach in “AI-free” classes, while others would teach some classes “with” AI and others “without.”
But above all, given the persistent disagreement on whether, all things considered, AI should be deployed in education, bilingualism would allow both hypotheses to be empirically tested on the same learner populations, rather than on separate cohorts subjected to the two regimes, by declining the comparison according to level, from primary school to university.
It also allows us to escape the all-or-nothing alternative except through current hybrid systems, combining traditional methods and low doses of AI in the same courses, a cautious approach, but one that does not allow progress on the unresolved theoretical and practical questions posed by the irruption of AI in education.
Author Bio: Daniel Andler is Professor Emeritus of Philosophy at Sorbonne University
