Education Policy / AI / Education Reform / Global Policy
AI in Education
How governments can learn from each other’s policies on AI in schools
of university students in a 16-country 2024 survey said they used AI in their studies (Digital Education Council)
Executive Summary
Michael Muthukrishna studies how AI could improve education and why technology alone is not enough. His background paper for UNDP’s Human Development Report Office, “AI Can Revolutionize Education but Technology Is Not Enough and Most Countries Are Not Ready,” is cited in the 2025 Human Development Report. A 2025 policy paper with five LSE co-authors argues that AI could boost learning and narrow gaps, but only inside systems that support teachers, build infrastructure and fit local culture.
In A Theory of Everyone, Muthukrishna argues for far more borrowing of policies that work elsewhere. He calls this policy plagiarism, and this page applies it to AI in schools.
Who's behind this
Why governments struggle to keep up
Muthukrishna and his co-authors write that “innovations in AI are moving faster than innovations in policy.”
If these barriers are not overcome, the paper argues, AI cannot narrow the development gaps between countries.
How policy plagiarism works
Policy plagiarism means borrowing a policy that has worked elsewhere, understanding why it worked, and adapting it to local institutions and values. In A Theory of Everyone, Muthukrishna points to Estonia, which ranked top outside East Asia in PISA, the OECD’s tests for 15-year-olds, while spending less per pupil than the OECD average. He attributes this to reforms that borrowed the best educational practices from around the world.
Muthukrishna and his co-authors write that innovations succeed “not just because they work, but because they fit into a culture and evolve through social adoption.” CHP looks for policies that have worked elsewhere, connects the people who can adapt them, and breaks the approach into four steps.
1. Scan
Map how other countries govern AI in schools and train teachers.
2. Filter
Choose policies that work elsewhere and judge how well they fit local conditions.
3. Adapt
Redesign each idea for local culture and government, with the people it affects.
4. Embed
Turn principles into practical rules for procurement, training, assessment and infrastructure.
Promises and risks of AI in schools
The 2025 paper judges AI by the capabilities it expands, meaning the real freedoms students and teachers have to learn, teach and take part in society.
Lessons from earlier education technology
The paper draws its lessons from three earlier programs.
Three programs, and what came next
Estonia’s Tiger Leap, launched in 1996, connected every school to the internet by 2001 and built coding and digital problem-solving into everyday learning
One Laptop per Child gave out cheap laptops without enough support for teachers, curricula or infrastructure, and billions were spent with “no educational gain to show for it”
Uruguay’s Plan Ceibal, an exception among One Laptop per Child projects, treated the laptop as one part of a long-term program that built infrastructure, trained teachers and supplied local content
Estonia’s AI Leap 2025 strategy aims to bring personalized learning systems, open-source tools for teachers and AI literacy into the core curriculum
Three missteps to avoid
Technology alone
One Laptop per Child assumed that access to laptops would transform education. Estonia and Uruguay introduced computers alongside local curricula and teacher training developed with teachers.
Weak infrastructure and support
Laptops often reached schools without stable electricity, internet access or maintenance, and some broke down with no one local to repair them.
No local adaptation
Engineers at MIT designed the laptops with little input from educators in the countries receiving them, and the preloaded content often did not match local curricula or languages.
Michael Muthukrishna
A Theory of Everyone (2023)
Five recommendations for AI in schools
The 2025 policy paper makes five recommendations.
Set goals for what AI should do for people, such as foundational skills for every child or more teacher time for mentoring.
Treat devices, connectivity and AI access as part of the right to education.
Train teachers to use AI, and teach students how it works and where it fails.
Protect student data, make AI decisions explainable, check high-stakes systems for bias, and let students appeal.
Involve teachers, students and parents, and pilot new tools in some schools before national rollout, a strategy the paper calls “fail locally, learn globally.”
The paper warns that AI could widen inequalities if the best tools reach only well-funded schools and wealthy countries.
Published by Center for Human Progress (US) Inc., a 501(c)(3) organization