For several decades, Western schools have been grappling with a persistent paradox. Never before have discussions about individualized learning, pedagogical support, and respecting students' learning paces been so prevalent; yet never before has a sense of ineffectiveness and discouragement been so widely shared by teachers, students, and families. Successive reforms have multiplied programs, modified assessment methods, and introduced new digital tools, without truly questioning the fundamental architecture of the school system. One of the central tensions remains largely unaddressed: the conflict between the progression of knowledge, the time required for its repetition, and the methods of its assessment. In practice, these three dimensions are conflated within the same space, that of the synchronized classroom , producing mechanisms of pedagogical and social confinement (the Pygmalion effect) [1] See source : rapid formation of stable academic reputations, early internalization of failure or excellence, rigidification of learning trajectories, and progressive reduction of opportunities for cognitive mobility within the class group. Students are assigned to implicit categories very early on, often interpreted as lasting traits, while teachers find themselves forced to manage a structural heterogeneity that exceeds their room for maneuver.
In this context, the emergence of artificial intelligence systems has reignited the educational debate, often in an excessive manner, oscillating between promises of radical transformation and principled rejection. The aim of this article is neither to succumb to technological enthusiasm nor to defend a weakened status quo, but to examine the possibility of a limited use of AI as a tool for repetition, serving to create a clearer separation between teaching, practice, and certification. It is about envisioning an innovative pedagogy not through the accumulation of tools, but through a reconfiguration of the fundamental functions of the school.
State of the'art on the'AI in education
Recent work on artificial intelligence in education converges on one essential point: AI does not, in itself, constitute a pedagogical model, but rather a set of technologies whose effects depend closely on the institutional, didactic, and ethical frameworks within which they are deployed. In France, publications from the Ministry of National Education, the CNIL (National Commission for Information Technology and Civil Liberties), and the Senate all emphasize the need to preserve the central role of the teacher, the protection of student data, and the non-automatability of certification decisions.
At the international level, UNESCO and OECD recommendations adopt a similar stance. They emphasize both the potential of AI to support learning and the risks associated with poorly designed algorithmic personalization, including the illusion of control , cognitive dependence , and the opacity of assessment criteria . These texts promote a human-centered approach , in which AI is conceived as an assistance tool and not as a judgmental authority. At the same time, critical literature, notably from the Council of Europe, warns against the silent transformations of educational practices brought about by algorithmic systems, particularly when the indicators produced by digital tools end up influencing the behavior of students, teachers, and institutions without any real explicit debate on the standards they impose. Finally, in the French public debate, contrasting educational voices express either strong concerns about the potential dehumanization of schools or radical criticism of AI, perceived as a factor in lowering standards. While these positions may not be based on empirical research, they nonetheless constitute an ideological backdrop that would be unwise to ignore.
This brief overview shows that, despite the abundance of frameworks and recommendations, the structural question of the relationship between progression, repetition, and certification remains largely under-theorized. It is at this level that the contribution proposed here is situated.
Synchronized Classroom and the Failure of Rehearsal
The central feature of contemporary schools remains the synchronized classroom, organized around a group of students of comparable age, progressing according to a common curriculum and schedule. This system rests on a fictional assumption: that of a sufficiently homogeneous learning pace to allow for effective group teaching. However, this assumption clashes with the reality of cognitive, social, and emotional differences, which teachers are constantly required to compensate for.
In this context, repetition, though essential to any lasting learning, becomes problematic. It is socially stigmatized, interpreted as a sign of delay or maladjustment, and rarely recognized as legitimate work. Students who need time are publicly exposed to their difficulties, while those who progress more quickly are forced to wait, often at the cost of gradual disengagement. Assessment, conceived as a moment of verification, then tends to solidify positions rather than reflect dynamics. These effects are not due to marginal malfunctions, but to the ordinary functioning of the system. By conflating the spaces of transmission, practice, and recognition, the synchronized classroom mechanically produces trajectories of confinement , both for students and teachers, who are reduced to managing an average pace that doesn't fully suit anyone.
Rehabilitating repetition as a space of'learning
All mastery practices, whether intellectual, artistic, or athletic, rely on long, often unspectacular phases of repetition, during which mistakes play a structuring role. School, as an institution of socialization, has progressively rendered these phases invisible or shameful by subordinating them to collective evaluative deadlines. Rehabilitating repetition requires recognizing its own space, distinct from the public stage of the classroom and from summative assessment. Such a space would allow students to engage in practice without every difficulty being immediately interpreted as a failure.
The aim would be to restore a right to trial and error, to experimentation and to slowness, a necessary condition for real progress.
This separation, however, poses a major practical problem on a large scale: how to offer each student individualized support for repetition without unduly increasing the workload of teachers? It is in this context that a limited use of artificial intelligence can be considered. AI is not conceived here as a substitute for the teacher, nor as an autonomous assessment tool, but as a repetition agent, strictly aligned with the content and progression defined by the teacher. It acts as an interactive platform allowing the student to practice, receive immediate feedback, and measure their progress without social exposure. In this configuration, the primary relationship is that between the student and the repetition agent. AI offers a space free from symbolic judgment, where the work in progress can be valued for its own sake. It also makes visible elements that are generally invisible, such as the time invested, perseverance, the nature of errors, and the diversity of strategies used. The data produced by this repetition space does not constitute a judgment. These records constitute traces, made available to the teacher, who remains solely responsible for their interpretation. AI sheds light on the student's actual work, without replacing pedagogical judgment.
Teaching a mobile cohort and assuming the'asynchronous
The separation of teaching and repetition profoundly transforms the teacher's role. The teacher becomes fully responsible for structuring knowledge, ensuring the epistemological coherence of the proposed chapters, and interpreting learning trajectories. No longer constrained to advance an entire group at an artificially unified pace, the teacher accepts a form of institutional asynchrony . Students progress at different rates, some rapidly advancing through several stages, others requiring more repetition cycles. This variability is no longer interpreted as an ontological hierarchy, but as a normal feature of the learning process. To avoid any arbitrary deviation, this asynchrony must be governed by explicit rules, particularly regarding the maximum duration of engagement in a given cycle. The decision to advance or redirect a student remains a human judgment, acknowledged and open to discussion, informed but not dictated by data derived from repetition. At the heart of this restructuring lies the certification function of schools and universities. The institution does not simply guide students; it publicly attests to socially recognized levels of mastery. This function, often euphemized, is nonetheless central and politically sensitive. The proposed model does not challenge certification; on the contrary, it reinforces its legitimacy by clearly distinguishing the time of learning from the moment of recognition. It also requires clarifying what is certified, at what level, and under what conditions. In doing so, it brings to light a question too often overlooked: that of the horizon of knowledge. Deciding that an individual has reached a relevant threshold of competence is never a purely technical observation. It is a normative decision, based on epistemological and social choices. AI can contribute to documenting learning pathways, but it cannot define this horizon without unduly replacing collective judgment.
What it means
Let us take two examples, which are not intended as prescriptive models, but aim to make a plausible pedagogical architecture intelligible.
1- In the first case, a cohort of students aged 14 to 16 is monitored in mathematics. While maintaining the traditional classroom model, but disregarding the concept of a "cohort level" (e.g., 10th, 11th grade) and avoiding the logic of automatic progression by age and school calendar—which leads students to advance collectively regardless of their actual level of mastery—the teacher spends six hours a week in class with the students, teaching in a lecture-style format. The lesson is entirely devoid of exercises; it focuses solely on knowledge , questions , and progression. Students, through discussion, work together to understand the underlying mechanisms. The teacher assigns no homework between lessons; instead, they follow a progression adapted not to the students' pace, but to the time they deem necessary for a given group to grasp the concepts. The learning calendar follows a common, national framework that is not based on age or grade level, but rather on an internal progression in the acquisition of knowledge. If students have acquired and validated a level, whether after one month or two years, they advance to the next level.
Progress is validated using a certification-by-repetition model: each student, at home, is supported by an AI agent entirely controlled by the course teacher. Repetition is neither forced nor required; however, the AI agent is capable of generating a report on the student, evaluating not only their ability to manipulate mathematical objects during exercises, but also their commitment to success, their tenacity, and their desire to succeed. This report is sent to the teacher and is an integral part of the assessment. AI-assisted repetition, which simply replicates the former role of teaching assistants at university, offers the opportunity to break free from the constraints imposed by the classroom, allowing children to showcase their work and effort and to demonstrate—in a quantifiable and honest way—their diligence. Demotivation and disinterest should be interpreted for what they are: a child's inability to adapt to the science being presented to them.
2- In the second case, a cohort of students aged 19 to 23 is monitored in a literature course. The same support system is used: the AI, controlled by the professor, acts as a "repetition agent" that provides students with exercises and the opportunity to self-assess and measure their progress. These exercises do not aim to automate literary interpretation, but rather to progressively develop the skills necessary for it: identifying argumentative structures, contextualizing works, comparing critical readings, practicing problematization, reformulating, and methodically constructing a commentary or essay. The professor receives a report from the AI on the student, again measuring their engagement, diligence, tenacity, and success. The final assessment—which could be an essay, a commentary, or an oral presentation—is therefore undertaken with full awareness of the effort involved. It certifies a framework, content, and methodology. Freed from the burden of exercises and homework, the Professor can fully dedicate his time to teaching his subject. The class no longer adapts to the weakest students, which discourages the high achievers. Instead, it fosters objective progress: AI becomes the means by which those who genuinely wish to can adapt. It is also a way for the best students to prove their worth.
The repetition model is by far the most effective for the real progress of pupils and students. Proof of this lies in the fact that the private tutoring market has never been so successful . But what is a private tutor today if not a personal repetition tool? Currently, this parallel education market is a factor of unfair differentiation , producing inequality and resentment. The existence of AI agents, controlled by teachers and made available to everyone by the institution, would initiate a desirable virtuous cycle.
Conclusion(s)
The inherited school model, based on synchronized learning paths and a blurring of pedagogical roles, is now showing its limitations. The responses offered so far oscillate between nostalgic conservatism and superficial innovations, without any structural transformation. By proposing a clear separation between teaching, repetition, and certification, and by considering the use of AI as a tool for repetition rather than as a means of judgment, it becomes possible to loosen the mechanisms of confinement that weigh on both students and teachers. Such a transformation is not a technological solution, but a deliberate institutional and political choice. Artificial intelligence is neither a fatal threat nor a panacea. It can, however, reveal the choices that schools are already making implicitly. As such, it compels us to rethink what we teach, what we assess, and what we certify. Perhaps it is in this demand for clarity that true pedagogical innovation lies.
Bibliography
Institutional frameworks and reports – France
Ministry of National Education (France). 2025. "Framework for the use of AI in education". Published in June 2025 (presentation page).
https://www.education.gouv.fr/cadre-d-usage-de-l-ia-en-education-450647
French Ministry of National Education. 2025. AI in education: framework for use. PDF document (version distributed by the Créteil Academy).
https://www.ac-creteil.fr/sites/ac_creteil/files/2025-06/l-ia-en-education—cadre-d-usage-36099.pdf
CNIL (National Commission for Information Technology and Civil Liberties). 2025. "Data controller: how to implement AI systems in education?". June 20, 2025.
https://www.cnil.fr/fr/education-mise-en-place-systeme-ia
CNIL (National Commission for Information Technology and Civil Liberties). 2025. "Teachers: how to use an AI system in the school setting?". June 20, 2025.
https://www.cnil.fr/fr/enseignant-usage-systeme-ia
Senate (Senate Delegation for Foresight). 2024. AI and Education. Information Report No. 101 (2024-2025), rapporteurs Christian Bruyen and Bernard Fialaire. Filed on October 30, 2024.
https://www.senat.fr/rap/r24-101/r24-101_mono.html
International organisations
OECD (Organization for Economic Co-operation and Development). 2026. OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. OECD Publishing, Paris. 19 January 2026. DOI.
https://doi.org/10.1787/062a7394-en
OECD (Organization for Economic Co-operation and Development). 2026. OECD Digital Education Outlook 2026 – PDF.
UNESCO. 2023. “Guidance for generative AI in education and research.” Presentation article. September 7, 2023; Last updated: January 16, 2026.
https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research
UNESCO. 2023. Guidance for generative AI in education and research. UNESCO (UNESDOC – notice).
https://unesdoc.unesco.org/ark:/48223/pf0000386693
UNESCO. 2023. Guidance for generative AI in education and research. PDF (UNESDOC).
https://unesdoc.unesco.org/ark:/48223/pf0000386693/pdf/386693eng.pdf.multi
Rights, governance, grounded critique
Council of Europe. nd (editor page). Artificial intelligence and education: A critical view through the lens of human rights, democracy and the rule of law. Council of Europe Publishing (Book.coe.int – catalog page).
Council of Europe (institutional website). and “Artificial intelligence and education” (presentation of the report and framing by the values of the Council of Europe).
https://www.coe.int/en/web/education/-/artificial-intelligence-and-education-2
Works / Summaries
Holmes, Wayne; Bialik, Maya; Fadel, Charles. 2019. Artificial Intelligence in Education: Promises and Implications for Teaching and Learning. Center for Curriculum Redesign. (UCL deposit page).
https://discovery.ucl.ac.uk/id/eprint/10139722
Holmes, Wayne; Bialik, Maya; Fadel, Charles. 2019. Artificial Intelligence in Education: Promises and Implications for Teaching and Learning. PDF extract (Center for Curriculum Redesign).
https://curriculumredesign.org/wp-content/uploads/AIED-Book-Excerpt-CCR.pdf
AI policy context (France) – historical milestone
Villani, Cédric (ed.). 2018. Giving meaning to artificial intelligence: towards a national and European strategy. Report (PDF).
Ministry of Higher Education and Research (France). 2018. “Report by Cédric Villani: giving meaning to artificial intelligence (AI)”. November 28, 2018 (presentation page).
[French public debate (contrasting positions, not “evidence” but points of reference)]
Meirieu, Philippe. 2025. “With AI, it is the political model of our school that is at stake.” Le Monde, September 16, 2025. (Article).
Brighelli, Jean-Paul. 2025. “AI, or widespread mediocrity.” Causeur, July 21, 2025.