Published March 22, 2026 | AI in Education Weekly | Issue #5
Last December, education researcher Thomas Hatch visited Suzhou Experimental Primary School — a school old enough that John Dewey walked its hallways in 1919. He watched teachers use AI to generate personalized learning materials, help students explore artistic styles, and adapt content in real time to individual needs. What he heard from one teacher stopped him:
“AI is a museum that helps us find inspiration, not a print shop that gives finished products.”
That line has been living in my head ever since I read it.
Because right now, in classrooms from Beijing to Singapore to my own AP World History room in Shanghai, that’s exactly the tension we’re all navigating: Are we building museums? Or are we running print shops?
The OECD Has Data — And It’s a Warning
Earlier this year, the OECD released its Digital Education Outlook 2026, the most comprehensive international study yet on generative AI in classrooms. The findings are honest in a way that a lot of the AI-in-education conversation hasn’t been.
Here’s the uncomfortable part: students who used general-purpose AI tools produced higher-quality work in the short term — but that advantage disappeared when they were tested without the tool. The output improved. The learning didn’t. That’s something that we have all been made aware of, and if we’re not careful, it could just get worse and worse.
The OECD calls it “metacognitive laziness.” When students can outsource the thinking, many of them do. They hand the hard cognitive work to the machine and walk away with something that looks like learning. It isn’t.
This isn’t an argument against AI. The same report found that when AI tools are designed with pedagogical intent and embedded in sound teaching strategies, they genuinely enhance critical thinking, creativity, and collaboration. The difference isn’t the tool — it’s what the teacher does with it.
As the OECD puts it: “How GenAI is designed and used matters more than whether it is used at all.”
That should be tattooed on every professional development slide in every international school on earth right now.
Asia Is Going All-In (And Not Just China)
Last week I wrote about China’s mandatory AI curriculum, which reached every primary and secondary student in the country last September. But China isn’t alone in making a decisive move.
India is rolling out a compulsory AI curriculum starting from Grade 3 — for all schools — beginning in the 2026-27 academic year. The Council for the Indian School Certificate Examinations (CISCE) already added robotics and AI to its curriculum in 2025-26. That’s hundreds of millions of students moving toward structured AI literacy, simultaneously, across the world’s two most populous countries.
Malaysia is tying its AI education push directly to economic goals: the digital economy is projected to contribute 25.5% of Malaysia’s GDP by 2025. Its Digital Education Policy isn’t framed as a tech experiment — it’s framed as national competitiveness infrastructure.
Singapore’s AI literacy initiative, built on top of its Student Learning Space platform and SkillsFuture framework, is aiming to have teacher training at all levels complete by 2026. Singapore has been doing this quietly and methodically, which is classic Singapore.
What’s striking when you look at the region together is the consistency of the underlying concern: governments that are serious about AI competitiveness have stopped waiting for organic adoption and started building it into the system. The international schools embedded in these countries are in a fascinating position — they’re navigating Western curriculum frameworks (IB, AP, Cambridge) inside national contexts that are sprinting toward something those frameworks weren’t designed for.
What We’re Not Teaching — Yet
Here’s the part that most AI-in-education conversations skip: power.
Leon Furze, author of the just-released open-access book Teaching AI Ethics, has spent years building classroom frameworks for the ethical dimensions of AI. His nine-topic series covers the expected ground — bias, privacy, academic integrity — but the one that hits hardest for me is the last one: power.
Furze’s argument is that the ethical concerns of AI — from bias and labor displacement to environmental cost — don’t exist in isolation. They coalesce to reinforce existing societal power structures. The infrastructure of AI raises questions about who controls the training data, who makes the architectural decisions, who profits, and who doesn’t. His concept of hegemony isn’t academic jargon here — it’s a practical question: when your students use an AI tool, whose values are embedded in it? Whose voices were excluded from the data it learned from?
For a history teacher, that question is not new. We’ve been asking it about primary sources for decades. A telegram from a colonial administrator tells one story. The silences in the archive tell another. AI-generated content has the same structure — it’s made from sources that reflect the power of whoever had the means and authority to create and preserve them.
The good news, as Furze notes, is that quality instruction on AI ethics doesn’t require a specialized AI literacy class. It requires doing what good humanities and social science teachers already do: interrogating whose perspective is centered, who benefits, and what gets left out.
What This Looks Like in My Classroom
I’ll be specific, because I think the “here’s how AI can be used in a history class” conversation often stays too abstract.
Right now, the highest-leverage thing I do with AI in my AP World History class isn’t using it to generate content — it’s using AI’s limitations as a teaching tool.
I give students an AI-generated historical analysis of something we’ve already studied. Last month it was a summary of the causes and consequences of the Mongol Empire’s expansion. The AI’s version was fluent, organized, and missing about half of what mattered. It flattened the diversity of perspectives across the regions affected. It gave disproportionate weight to the sources most available in English. It hallucinated one citation.
Students had to find all of it. They came in the next day with annotations, corrections, and — this is the part I didn’t fully anticipate — frustration. One student said: “It sounds so confident. How are you supposed to know when it’s wrong?”
That question is the lesson. That’s the media literacy muscle Colleen Kenny is writing about in her LinkedIn piece on content chaos. Kenny — who teaches at NYU Tisch — argues that media trust has collapsed to 28% (Gallup) and that AI-generated content indistinguishable from authentic sources is accelerating that collapse. Her case: multilayered media literacy — cognitive, emotional, and embodied — is now democracy’s most urgent educational task.
In a history classroom, the vocabulary for this already exists: sourcing, corroboration, contextualization. We extend it to AI, and we’re already doing the work.
The Suzhou teacher’s museum metaphor keeps coming back to me here, too. I want my students to use AI the way you use a museum — to encounter ideas, challenge their assumptions, find unexpected connections. Not to walk out with a souvenir that someone else made.
For International School Teachers: A Practical Frame
The OECD report, Furze’s ethics framework, and the Suzhou observations all converge on the same practical point: the teacher’s pedagogical intention is the variable that matters most.
A few things that have been working in my room, and that I’ve heard work elsewhere:
Use AI’s errors intentionally. Assign AI-generated content not as a model but as a source to critique. Students learn more from finding what the AI got wrong than from imitating what it got right.
Ask the power question explicitly. Before using any AI tool with students, spend five minutes on: Who made this? What data was it trained on? Whose perspectives are likely underrepresented? This doesn’t take a separate ethics unit — it takes five minutes and a habit of mind.
Design for the exam, not the shortcut. If your assessment can be completed by AI, rethink the assessment. The OECD data is clear: AI-generated outputs don’t survive into exam performance. Build tasks that require synthesis, oral defense, or documented process — and AI becomes a scaffold, not a substitute.
Name the metacognitive risk out loud. Tell students: the AI will make you look smarter than you are in the short term, and it will make you worse at thinking in the long term, if you let it. That’s not a moral lecture. That’s just the data.
The Bigger Question
I keep coming back to Thomas Hatch’s observation in Suzhou. The school he visited wasn’t trying to outpace technology or eliminate it. The teachers were asking a harder question: What does education actually need?
Not what technology can provide. What education — human formation, genuine understanding, the ability to think in the presence of ambiguity — actually needs.
That’s the frame I want for international schools navigating this moment. Not “how do we integrate AI” but “how do we make sure AI serves what we’re actually here to do?”
The museum doesn’t print anything. It asks you to think.
Sources and further reading:
• Teaching AI Ethics 2026: Power — Leon Furze
• Teaching AI Ethics: Free Open Access eBook — Leon Furze
• Shockwaves and Innovations: How Nations Worldwide Are Approaching AI in Education — CRPE
• India to Introduce AI Curriculum in All Schools by 2026 — TechWire Asia
• Schools in China Are Making AI Part of the Curriculum — NPR
• Learning in the Age of AI: How China’s International Schools Are Adapting — AmCham China
