Moral
Red Lines White Paper
Moral red line 03 Human good · intellectual formation

Education, Intellectual Formation, and Cognitive Offloading

Learning grows through attention, effort, questions, and discovery. Educational AI should leave students more able to understand, remember, explain, and act for themselves.

What can the learner understand and explain when the assistant is put away?

The moral boundary

Educational systems must not normalize AI use that substitutes machine-generated performance for the intellectual capacities learners are supposed to acquire or demonstrate. The boundary is crossed when AI routinely performs the reading, recall, reasoning, problem-solving, writing, correction, or explanation that constitutes the student's formative task, especially before foundational skills are secure or during assessment of independent competence. This protects learners' ability to understand, remember, explain, and act with their own judgment.

The pro-human direction

Design and use AI to widen access, support practice, offer feedback, and prompt reflection. Preserve formative intellectual effort, teachers' judgment, parents' educational responsibility, and protected opportunities for independent learning and assessment.

Students' reported AI use

46%

PISA 2025, published in September 2026, reported that 46% of students across OECD countries used AI chatbots at least weekly to help them learn. Its comparisons between AI use and attainment are observational and reflect students' differing circumstances.

When better output hides weaker learning

A child struggles with a problem. A teacher asks a helpful question. The child tries again and finally understands. That moment contains the promise of education: growing in the ability to think and act with one’s own judgment.

AI can supply an explanation or solution in seconds. The important question is what happens to the learner along the way. A polished assignment may conceal missed opportunities to read closely, remember, reason, and correct a mistake.

The foundations already need care

The September publication of PISA 2025 reported another decline between 2022 and 2025: 14 points in reading and nine in mathematics across comparable OECD countries. Reading declined significantly in 56 of 74 education systems with comparable results since 2018. This longer decline predates widespread generative-AI adoption and has multiple possible causes, including disrupted schooling, inequality, changing reading habits, and digital distraction.12

PISA also found weekly AI use among 46% of students across the OECD. Students using it for activities such as drafting, summarizing, and research scored around 20 points lower in science. This observational association includes differences among students and their circumstances. It calls for closer examination of how assistance affects learning.3

A National Literacy Trust survey of 40,543 young people found widespread weekly use and roughly a quarter sometimes copying AI homework outputs. Jonathan Haidt describes the concern as cognitive surrender: repeated opportunities to hand over thinking while important capacities are still forming.412

Learning after the tool is removed

Experiments can examine this question more directly. Hamsa Bastani and colleagues studied nearly 1,000 high-school mathematics students. Students using an unguarded AI assistant performed 17% worse on a later unaided test than the comparison group. A tutor configured to offer learning-oriented support largely avoided that loss.5

The OECD Digital Education Outlook draws a practical lesson: general-purpose AI can improve completed work while learning remains weak; systems designed around sound pedagogy can strengthen learning. Teachers need evidence of what students retain and can do independently.6

Moral foundation

Education forms the whole person

A teacher offers a question, waits through uncertainty, and helps a student try again. Parents sustain the patience and habits that make such learning possible. Gravissimum Educationis understands education as the formation of the whole person and gives parents a primary educational responsibility. Understanding grows alongside freedom, character, and the ability to live with others.7

Magnifica Humanitas connects education with attention, memory, perseverance, and independent judgment. Its account gives teachers and families a place in deciding how technology enters the classroom. Students need opportunities to defend an answer, admit uncertainty, listen, and change their minds.8

Speaking at UNESCO on 25 September, Leo XIV called for technology shaped around young people's education, relationships, and freedom to think and discern. That freedom develops through the work a learner gradually makes their own.9

01

Freedom grows through learning

Attention, memory, reasoning, and explanation give learners the ability to judge and act for themselves.

02

Patience and effort matter

Trying, making mistakes, recalling, and revising help knowledge become the student's own.

03

Families and teachers participate

Those responsible for a child's formation need a genuine voice in the purposes and limits of educational technology.

From principle to practice

AI that leaves the learner stronger

Good tutoring asks a question, offers a hint, invites recall, and lets the learner attempt an answer. AI can widen access, adapt practice, assist students with disabilities, and give timely feedback. Schools should assess whether those benefits endure when students work independently.

A September preprint reports a preregistered experiment with 704 adults practising fraction arithmetic. Feedback about how much reasoning participants had handed to an assistant reduced requests for complete answers and improved immediate unaided performance. The study concerns a short online task and offers a direction for further testing.10

Teachers, families, and students should take part in adoption decisions. UNESCO's September ministerial statement calls for public accountability, age-appropriate AI, teachers' judgment, and education as a common good. Protecting these duties includes AI-free learning and assessment, children's data protection, and withdrawal of products that weaken retained learning.11

Schools

Assess retained understanding. Use oral explanation, drafts, recall, and protected independent work.

Designers

Build tutors around questions, hints, practice, and reflection. Test learning after assistance is removed.

Families and teachers

Match assistance to the learner's stage, preserve foundational skills, and choose tools together.

Minimum conditions

  • Secure foundational skills before routine offloading
  • Questions and hints before finished answers
  • Evidence of retained and independent learning
  • Accessibility and age-appropriate support
  • Protected AI-free practice and assessment
  • Teacher and family participation with children's data protection
Selected sources Every source is drawn from the white paper and opens the primary document or publication page where available. View sources
  1. 1
    PISA 2025 Results (Volume I): Future-Ready Students; student performance

    OECD · 8 September 2026 · International assessment report; chapter 2

    Government / regulatorInstitutional report

    The long-term decline precedes widespread generative-AI adoption and has multiple possible causes.

  2. 2
    Trends in reading scores over three PISA cycles (2018–2025)

    OECD · 8 September 2026 · PISA 2025 annex A1

    Government / regulatorInstitutional report

  3. 3
    PISA 2025 Results (Volume I): Future-Ready Students; AI use and student performance

    OECD · 8 September 2026 · Observational student comparisons

    Government / regulatorInstitutional report

  4. 4
    Young people teachers and parents' use of AI to support literacy in 2026

    National Literacy Trust · 18 August 2026 · Survey of 40

    Civil society / research instituteInstitutional report

  5. 5
    Generative AI without guardrails can harm learning: Evidence from high school mathematics

    Hamsa Bastani et al. · 2025 · PNAS 122 no. 26; e2422633122; randomized experiment

    AcademicPeer-reviewed

  6. 6
    OECD Digital Education Outlook 2026

    OECD · 19 January 2026 · Institutional synthesis

    Government / regulatorInstitutional report

  7. 7
    Gravissimum Educationis, §§1, 3

    Second Vatican Council · 28 October 1965 · Declaration on Christian education

    ChurchOfficial primary source

  8. 8
    Magnifica Humanitas, §§139–147

    Leo XIV · 15 May 2026 · Encyclical

    ChurchOfficial primary source

  9. 9
    Address at UNESCO headquarters in Paris

    Leo XIV · 25 September 2026 · Papal address

    ChurchOfficial primary source

  10. 10
    Designing Against Deskilling: Metacognitive Feedback Reduces Cognitive Offloading to LLM Assistants

    Sebastian Maier et al. · 17 September 2026 · Preprint; preregistered online experiment with 704 adults

    AcademicPreprint / emerging

  11. 11
    Education Ministers call for education to remain a common good in the age of AI

    UNESCO · 9 September 2026; statement presented 8 September · Ministerial statement

    Government / regulatorOfficial primary source

  12. 12
    Should We Ban AI for Kids?

    Jonathan Haidt and David Brooks · 4 October 2026 · The Atlantic podcast; expert discussion

    JournalismExpert commentary / forecast