🚨AI in schools: inclusion tool or machine for deepening inequalities?
🔍💡 AI and learning: early use is already here (observed from Year 6), but support is lacking. The report draws the line between AI that includes and that which dispossesses. 🧠
📌This report documents, with supporting figures and hearings, a reality already established: young people are massively using AI, often out of the sight of adults, and at an increasingly younger age. It provides a framework to distinguish between uses that genuinely help (accessibility, remediation, differentiation) and those that impoverish (cognitive delegation, worsening inequalities). For a support professional, it is a positioning tool: it allows for objectifying what is already observed and anchoring a discourse that is neither technophobic nor naive. The sections on disability, non-native speakers, and early school dropout are directly transposable to medico-social practices.
Source:📒 Creation, dissemination and acquisition of knowledge: how artificial intelligence is transforming our education and our culture✍️ Mr. Roger Chudeau (chair of the mission) and Ms. Céline Calvez (rapporteur) — Committee on Cultural Affairs and Education, National Assembly - Information report submitted in accordance with Article 145 of the Rules by the Committee on Cultural Affairs and Education, concluding the work of the information mission📜🔗LINK to the source
1️⃣ ANALYTICAL SUMMARY
A massive adoption that is taking the institution by storm. Generative AI has spread at an unprecedented speed: 48% of the French population will use it in 2025 according to Crédoc, and usage is almost universal among 16-25 year olds. Among students, appropriation is becoming increasingly early — a usage rate once observed in secondary school is now seen as early as Year 6 — and is primarily taking place outside the classroom, without support. The report rejects a defensive stance: it acknowledges legitimate criticisms (decreased cognitive engagement, delegation of intellectual tasks) but considers that schools cannot ignore or sustainably prohibit a tool that has already entered the daily lives of young people.
A success conditioned on training and equity. The document carefully distinguishes between public conversational agents — which provide answers and flatter the user — and educational ‘Socratic’ AIs, designed to support learning. It identifies concrete contributions for vulnerable groups (disabilities, non-native speakers, learning difficulties) while alerting to a major risk: without support, AI “enriches the rich and impoverishes the poor,” creating a two-speed system. Its 26 recommendations focus on training staff (including AESH), education on AI from school onwards, maintaining learning time without AI, and sovereign and equitable equipment.
2️⃣ KEY POINTS OF THE DOCUMENT
1️⃣ Youth usage is widespread and early. 89% of 16-25 year olds have already used AI (compared to 43% of the general population), and usage is now descending to the CM2 level; in higher education, 92% of students use it and nearly one in two would find it difficult to do without it (p. 119).
2️⃣ The distinction between general AI / educational AI is structuring. Socratic AIs ask guided questions, require justification of answers, and stimulate cognitive engagement, whereas mainstream chatbots prioritise immediate responses and user complacency (p. 122-123).
3️⃣ AI is a concrete lever for accessibility for disabilities. Transformation from written to oral for visual impairments, rewriting in easy-to-read language and pictograms for cognitive disorders, automatic subtitling for hearing impairments, formatting adaptations for dyslexia — hence the recommendation to also train AESH (recommendation no. 17, p. 126-128).
4️⃣ The risk of "cognitive dispossession" is clearly posed. The systematic delegation of tasks (synthesis, writing) to AI threatens skills that students have not yet acquired; hence the necessity to maintain "AI-free" times and an "educational bilingualism" (knowing how to do with and without) (p. 139-140).
5️⃣ AI can exacerbate inequalities. A typology distinguishes between "reflective engaged", "occasional legalists" and "opportunistic students": the latter, often from working-class backgrounds, delegate all work without understanding, with a digital divide affecting access to effective and paid tools (p. 153-155).
3️⃣ ACTION TRACKS FOR LOCAL ACTORS
1️⃣ Map the actual uses of your audience before any normative discourse: the typology into three profiles (p. 154-155) is a directly usable tool in interviews or workshops to identify who delegates without understanding.
2️⃣ Utilise AI as a documentary accessibility tool : rewriting in plain language, summaries, adding pictograms, audio versions of materials — useful for allophone audiences or those with cognitive disabilities (p. 126-129), provided that the human element remains in the loop.
3️⃣ Use AI as a disinhibiting training space for individuals self-conscious about their language gaps: practice with an agent before exposing oneself to the gaze of a professional (usage praised in UPE2A, p. 128-129).
4️⃣ Establish times without AI in your support sessions to preserve basic skills (writing, summarising, memorisation), following the model of “mathematics without a calculator” (p. 140).
5️⃣ Integrate screen awareness into your practice : the report reminds us of the documented effects on attention, sleep, and sedentary behaviour, and encourages qualifying uses rather than demonising “screens” as a whole (p. 137-139).
6️⃣ Align your actions with the official framework : the “framework for the use of AI in education” (June 2025) and the Pix/Pix Junior pathways provide ready-to-use guidelines for building a solid professional stance (additional reference below). Identified unmet need: the report acknowledges the lack of robust studies on the cognitive effects of educational AIs themselves — caution therefore on unsubstantiated promises.
4️⃣ ADDITIONAL REFERENCES
🔍➕ For more information, see the articles referenced by "Health Practices"
- HAS & CNIL, Supporting the proper use of artificial intelligence systems in healthcare contexts (February 2026) — directly aimed at the health and medico-social sector, it clarifies the responsibilities of professionals and structures regarding AI systems. 🔗 https://www.cnil.fr/sites/default/files/2026-03/guide_has_cnil_recommandations_ia.pdf (URL verified, active)
- Ministry of National Education, AI in education: usage framework (June 2025) — the reference text that sets the conditions for responsible, secure, and ethical use of AI in schools, cited by the report itself. 🔗 https://www.education.gouv.fr/cadre-d-usage-de-l-ia-en-education-450647 (URL verified, active)
- CNIL, Development of AI systems: recommendations for compliance with GDPR (July 2025) — methodology to reconcile the use of AI and data protection, essential when handling data from vulnerable populations. 🔗 https://www.cnil.fr/fr/developpement-des-systemes-dia-les-recommandations-pour-respecter-le-rgpd (URL verified, active)
5️⃣ FREQUENTLY ASKED QUESTIONS (FAQ)
1️⃣ From what age does the report recommend the pedagogical use of AI in the classroom? From Year 6, with a structured use possible from the first level (recommendation no. 18, p. 174). The official usage framework allows it from Year 4.
2️⃣ What is the difference between a general AI and an educational AI? The general AI (ChatGPT, etc.) provides the answer and tends to flatter the user; the educational "Socratic" AI guides through questions, requires justification of answers, and stimulates cognitive effort (p. 122-123).
3️⃣ Does AI really make students "dumber"? The report nuances: the media-covered study on ChatGPT and "cognitive debt" focused on adults, a limited sample, and only immediate effects. The real risk of "cognitive spoliation" exists, but depends on usage (p. 139-140).
4️⃣ How can AI help a student with a disability? Depending on the disorder: oral reading of a text, rewriting in easy-to-read language, subtitling, formatting adaptation. These functions sometimes pre-existed, AI simplifies and enhances them (p. 126-127).
5️⃣ Can we detect if an assignment was done by AI? Not reliably: the ministry and the CNIL advise against tracing tools, deemed unreliable and infringing on confidentiality. The reflection should focus on how to assess, not on tracking (p. 143-144).
6️⃣ Will AI replace teachers and support staff? No. The report asserts that the usefulness of the profession is beyond doubt; AI could even strengthen the role of safeguarding the collective and metacognition (p. 141-142).
7️⃣ What is the main social risk highlighted? A "two-speed" system: those who are better off benefit from AI, while those who are less supported delegate without understanding. Education in AI is presented as a safeguard against this widening of inequalities (p. 153-155).
6️⃣ REWRITING IN EASY LANGUAGE
What the document says
Today, many young people use artificial intelligence. We call it AI.
AI is a tool on a computer or phone.
It can write texts and answer questions.
Almost 9 out of 10 young people already use it.
Often, they use it alone, without an adult to help them.
Sometimes, they are very young.
AI can help
AI can help people with disabilities.
It can read a text aloud.
It can write a text that is easier to understand.
It can add subtitles to a video.
AI can also help people who are learning French.
AI also has dangers
If we ask everything from AI, we stop learning.
We can lose the habit of writing or thinking.
This is serious for children who are still learning.
Therefore, we need to keep moments of work without AI.
AI does not help everyone in the same way
Some young people know how to use it well. They learn better.
Others copy answers without understanding.
Often, these are young people from poor families.
We need to help these young people. Otherwise, the gaps will grow.
Adults remain important
AI does not replace teachers.
It does not replace support staff.
Adults help to understand. AI alone is not enough.
7️⃣ CROSS-ANALYSIS — VALUES OF HEALTH PRACTICES
Literacy : the report explicitly values tools for adapting to levels of understanding (Easy Language, summaries, partial translation, audio), but reminds that language proficiency conditions the effectiveness of AI.
Empowerment : learner autonomy is a guiding thread (autonomous learning, ‘mentorship’), tempered by the risk of passivity and delegation.
Participation : the cited official usage framework comes from a wide consultation of the educational community; the report itself relies on numerous hearings of practitioners.
Community health : the collective dimension is strongly defended — AI must not confine to a triptych of ‘teacher-student-machine’ to the detriment of the class collective.
Ethics : biases, hallucinations, and AI's tendency to flatter the user are identified as technical limits to be aware of for informed use.
Human rights : accessibility for disabilities and non-native speakers is a central axis, linked to the goal of school inclusion in mainstream settings.
Intersectorality : the report articulates national education, higher education, local authorities, CNIL, and Arcom around the deployment of AI.
Partnership : models are mentioned (P2IA partnerships, DescolIA project, Pix platforms), without detailed methodological formalisation.
Fight against discrimination: the risk of worsening social inequalities is addressed in depth; the diversity of student profiles and specific educational needs is taken into account without judgement.
8️⃣ EVALUATION OF THE RELIABILITY OF THE RESOURCE
Scientific relevance — strong, with acknowledged methodological reservations. The report relies on current and varied sources (Crédoc 2026, Ifop 2025, IGESR June 2025, report Children and screens 2024, recent court decisions) and on numerous hearings. It demonstrates intellectual honesty by pointing out the limitations of certain studies it cites (small samples, short-term effects measured only) and acknowledging the lack of robust studies on the cognitive effects of educational AIs. It is a document of a political and forward-looking nature (parliamentary report carrying recommendations), and not an academic study: its recommendations are of an argued advocacy nature, not of opposable proof.
Operational relevance — high for framing, medium for tooling. Very useful for positioning, arguing and identifying uses; less directly tool-oriented (few check-lists or ready-to-use protocols), with concrete tools being referred to official frameworks (usage framework, Pix).
⚠️ Internal inconsistencies noted (reported, not corrected):
- The report itself points out a "material error" : the obligation to train in AI is included in block 4 of the primary school teachers' master's programme but is absent from that of secondary education, even though AI is more likely to be used there (p. 147-148).
- Terminological fluctuation : the regulatory text of January 2026 is referred to sometimes as "decree of 12 January 2026", sometimes via "annex 2 of the decree" (p. 147).
- Two adoption figures coexist for the general population in 2025: 48% (Crédoc, Digital Barometer 2026) and 43% (Ifop for Jedha) — distinct surveys, not contradictory, but not to be confused.
9️⃣ STRATEGIC HASHTAGS
#HealthPractices #AIinEducation #SchoolInclusion #DigitalLiteracy #MediaEducation #SpecialEducationalNeeds #EducationalEquity #CriticalThinking