Singapore pushes three-quarters of its teachers toward artificial intelligence. France lags at 14%. Between these two extremes, global education is shifting toward a societal choice that will determine whether AI liberates pedagogy or industrializes it.
The 41% of OECD teachers already using AI in 2024 are sketching two antagonistic futures for education. On one side, pedagogical augmentation that individualizes learning and frees time for human support. On the other, evaluation automation that standardizes practices and transforms schools into cognitive assembly lines.
The Essentials
- 41% of OECD teachers use AI in 2024, with massive gaps between Singapore (75%) and France (14%)
- Europe bans emotional recognition in classrooms but authorizes behavioral analysis “for pedagogical purposes”
- 67% of educational systems are testing automated evaluations, threatening formative assessment
- Deployment costs range from $50 per student (basic tools) to $2,000 per class (integrated systems)
Singapore Industrializes Personalization, Europe Hesitates
The gap between Singapore and France reveals two incompatible educational philosophies. Singapore has massively deployed AI since 2019 with its “AI for Students” program, training 32,000 teachers in adaptive learning tools. Result: 75% of them use AI assistants to differentiate learning pathways and generate personalized exercises.
Singapore’s approach bets on systemic efficiency. Each student has a cognitive profile fed by their interactions with learning platforms. AI adjusts exercise difficulty in real time, identifies conceptual gaps, and proposes targeted remediation. This mass personalization allows teachers to focus on supporting struggling students rather than preparing standardized lessons.
France resists with 14% adoption, primarily due to institutional mistrust and lack of training. The Ministry of National Education timidly tests correction assistants since September 2024, but without an overall strategy. This hesitation contrasts with the urgency of needs: 47% of French teachers report lacking time for individualized support, according to the TALIS 2024 survey.
Europe Bans Emotional Surveillance But Opens the Door to Behavioral Analysis
The European AI regulation, which entered force in August 2024, formally prohibits emotional recognition in educational establishments. This ban targets systems that analyze students’ facial expressions, posture, or voice to infer their emotional states. The European Union considers these technologies to violate human dignity and create an oppressive school environment.
But the regulation explicitly authorizes behavioral analysis “for legitimate pedagogical purposes.” This legal nuance opens a wide avenue for educational software publishers. ClassDojo, used by 95% of American elementary schools, already analyzes connection times, navigation patterns, and response patterns to adjust its pedagogical recommendations.
This soft surveillance concerns education researchers. Shoshana Zuboff, specialist in surveillance capitalism, warns of the transformation of schools into “behavioral laboratories” where every click, every hesitation, every error feeds predictive algorithms. Students become data sources before becoming learners.
Evaluation Automation Threatens Pedagogical Support
67% of OECD educational systems are experimenting with automated evaluations, according to TALIS 2024 data. These tools promise to eliminate subjectivity and accelerate grading. ETS, creator of the TOEFL, is deploying its e-rater system in 15 countries to automatically correct written productions. The algorithm analyzes syntax, vocabulary, and argumentative structure with 87% concordance with human evaluators.
This automation radically transforms the teaching profession. Gone are red-marked papers, handwritten annotations, and direct exchanges about errors. AI generates standardized feedback, categorizes difficulties, and proposes remediation exercises. The teacher becomes supervisor of a system that evaluates, diagnoses, and prescribes without them.
The risk of taylorization worries teachers’ unions. The FSU denounces an “intellectual proletarianization” that empties the profession of its relational dimension. Because formative assessment, the kind that accompanies daily learning, cannot be reduced to a grade or algorithmic diagnosis. It requires empathy, contextual adjustment, and pedagogical negotiation with the student.
The Finnish example illustrates this tension. Helsinki has been experimenting since 2023 with hybrid evaluations where AI pre-grades productions and the teacher validates results. This approach preserves professional judgment while saving time. But it requires in-depth training to avoid anchoring bias: the tendency to automatically validate algorithm suggestions.
Costs Fragment Access to Educational AI
The deployment of educational AI reveals major economic fractures. Basic content generation tools cost $50 per student per year. Khan Academy offers its Khanmigo AI tutor for $9 per month per family. These solutions remain accessible to Western middle classes.
But integrated systems explode budgets. A classroom equipped with behavioral sensors, interactive screens, and predictive algorithms costs between 15,000 and 50,000 euros. Only private schools and wealthy school districts can afford these investments. This digital divide worsens existing educational inequalities.
India is developing a frugal approach with its DIKSHA platform, which reaches 240 million students at a unit cost of $2 per year. This solution prioritizes adaptive content on smartphones rather than heavy equipment. But it limits functionality: no advanced behavioral analysis, no fine-grained personalization, no automated evaluation.
Germany Experiments With Democratic Co-Construction
Germany is testing a third path with its “AI & School” program launched in September 2024. Rather than imposing technological solutions, the federal ministry finances local experiments co-constructed by teachers, parents, and students. Each school defines its priorities: gain in administrative time, personalization of learning, or early detection of difficulties.
This bottom-up approach produces varied uses. The primary school in Hamburg uses AI to generate mathematics exercises adapted to each student’s cognitive profile. Munich’s gymnasium automates lesson planning and frees time for interdisciplinary projects. Dresden’s middle school develops an AI assistant that helps dyslexic students structure their writing.
These experiments avoid technocratic standardization while preserving pedagogical autonomy. But they slow down deployment and create territorial inequalities. Schools better equipped with digital skills fare better than those in rural or disadvantaged areas.
Toward Pedagogical Governance of Artificial Intelligence
UNESCO publishes in November 2024 its first recommendations for “ethical AI in education.” The report identifies four non-negotiable principles: algorithmic transparency, informed family consent, students’ right to error, and preservation of human pedagogical relationships.
These principles remain vague on their concrete implementation. How can we guarantee transparency of proprietary algorithms? How can we respect students’ right to digital oblivion when their data feeds predictive models? How can we preserve teachers’ pedagogical autonomy against increasingly prescriptive systems?
California has been experimenting since January 2024 with an “educational digital passport” that follows each student from kindergarten to university. This algorithmic file compiles performance, difficulties, learning preferences, and pedagogical recommendations. The stated objective: avoid disruptions in learning paths and personalize support. The implicit risk: lock students into predetermined trajectories from the earliest age.
The issue goes beyond technique to touch the very purpose of school. Training critical citizens capable of thinking for themselves, or optimizing learning pathways to maximize employability? Artificial intelligence reveals this fundamental tension and forces schools to choose their side.
Between Singapore, which industrializes personalization, and Europe, which hesitates out of caution, a third path is emerging: that of augmented AI that preserves the centrality of pedagogical relationships. This approach requires massive investments in training, binding ethical frameworks, and democratic governance of technological choices. The price to pay to prevent the school of the future from looking like a diploma factory piloted by algorithms.
Sources
- OECD TALIS 2024 - Teaching and Learning International Survey Results
- UNESCO - Recommendations for Ethical AI in Education, November 2024
- European Regulation on Artificial Intelligence, August 2024