I teach in Shanghai, which means I’m technically inside one system and watching another from the windows.
My students are at an American international school. Their relationship with AI is governed by our school’s policies — cautious, evolving, largely shaped by individual teacher decisions about what counts as authentic work. This is broadly true of American education right now: lots of variation, lots of uncertainty, policies that lag the technology by a reliable six to eighteen months.
Chinese public school students around us are operating under something different. The Ministry of Education made its move in 2024 and accelerated in 2025: AI in schools is not a district decision or a teacher’s call — it’s curriculum. Mandatory. National. By September 2025, every primary and secondary school in China was required to incorporate AI education. Beijing alone mandated a minimum of eight hours of AI instruction per academic year.
Vermont’s AI guidance for schools, released January 2026, threw that contrast into sharper relief.
The Vermont Framework
Vermont’s document is 50 pages. It’s careful. It reflects the state’s longstanding commitment to proficiency-based learning — the idea that education should produce demonstrable understanding, not seat time. The guidance recommends no AI chatbots at all for PreK through 2nd grade, curriculum-embedded AI only through 5th grade, education-specific chatbots for middle school, broader AI fluency by high school. Every step is graded and age-appropriate, oriented toward what the student needs rather than what the state needs.
The document frames AI as something that should amplify teacher judgment, not replace it. It connects explicitly to Vermont’s personalized learning goals — the idea that a student’s education should be tied to where they are, not to a standardized sequence. Educators lead implementation decisions. Student well-being is paramount. The approach is human-centered throughout.
It’s a thoughtful document. It’s also voluntary. No Vermont district is required to follow it.
The Comparison I’m Making
The comparison isn’t a clean one. Vermont is a state of roughly 650,000 people with a strong tradition of local control. China is a nation of 1.4 billion running a centralized education system as an instrument of national strategy. Comparing their AI policies is a little like comparing a town meeting to a five-year plan — they’re not trying to do the same thing, by the same means, for the same reasons.
That asymmetry is what makes the comparison productive.
Vermont’s guidance asks: how do we introduce AI in ways that protect what education is for — developing thinking, building knowledge, preparing students for complexity? The question is student-centered. The answer is cautious and permissive at the same time: here’s a framework, here are developmental guardrails, here are the values worth preserving. Go figure it out locally.
China’s guidance asks: how do we produce a generation capable of advancing China’s position in AI? The question is national-strategic. The answer is mandatory, sequenced, and aimed at output: AI-literate students who can contribute to the technology sector and the national goals behind it.
Same technology. Two different questions underneath the policy.
Why the Scale Difference Doesn’t Kill the Comparison
The comparison holds across the scale gap for a specific reason: Vermont’s model is close to the best version of what American AI education policy looks like when it’s done carefully. It’s not the average — the average U.S. state AI education policy is a paragraph in an acceptable use document. North Carolina issued AI literacy guidance in 2024 that Vermont built on. Vermont’s version is the rare one built around a developmental framework, with named stages of progression mapped to grade bands. Most other states haven’t published anything beyond general guidance. Vermont is the standard someone is trying to set. The rest of us are just hoping it turns out fine.
And we have evidence for what that hope produces. A 30-month study of 26,811 Chinese secondary students found generative AI use raised homework scores 18% and cut completion time 30%. Exam scores fell 20% within six months. On entrance exams that decide college placement, the gap grew to 18–24%, and the full hit only surfaced after about two years. Eighty percent of users showed the same pattern: fast completion, strong homework marks, and weaker performance once the tool was gone.
If you want to understand the philosophical gap between American and Chinese approaches to AI in school, Vermont gives you something real to compare.
And the gap is substantial.
China’s approach is producing AI fluency the way it produces STEM competency — at scale, through mandate, in service of national goals. The 2025 guidelines are tiered and developmental: basic AI literacy in primary school, critical thinking about machine learning in junior high, applied innovation and algorithm design in senior high. Real pedagogical thought went into the sequencing. The restrictions matter too — generative AI is prohibited for independent use in primary schools, and teachers are banned from using it as a substitute for core instruction. China isn’t just dropping ChatGPT into classrooms.
But the purpose of AI education in China is not individual flourishing. It’s national competitiveness. Students are being developed as AI-capable citizens — technically sophisticated, oriented toward production and innovation. The goal is workers and contributors.
Vermont, at its best, is trying to build something harder to measure: students who can use AI without being replaced by their own use of it. Students who retain the capacity to work through hard problems when the tool isn’t there. Students who understand the technology well enough to push back when it’s wrong.
The Failure Modes
Each model has a distinct failure mode.
Vermont’s: thoughtful guidance that doesn’t reach classrooms because teachers haven’t been trained to implement it. The data is clear — fewer than half of U.S. teachers have received any AI training from their schools, and only 29% have guidance on effective use. A 50-page framework without professional development is a document, not a policy.
China’s: students who can operate AI fluently but haven’t had to develop the cognitive habits that AI can bypass. Research on students and AI dependence is consistent across contexts — frequent AI use correlates with reduced engagement when the tool is removed, and students using generative AI for academic work sometimes show lower retention and critical engagement once it’s taken away. Mandating AI fluency and ensuring students can still think without it are different mandates.
Both countries are working against their own failure modes. Neither has fully resolved them.
What This Means for Teachers
The policy context matters less than what happens at the assignment level.
No national mandate guarantees students build real understanding. No thoughtful guidance document reaches a classroom a teacher hasn’t been prepared for. What reaches students is the sequence of tasks they’re asked to do, whether the difficult cognitive work is required before the tool is available, and whether the teacher has thought carefully about the difference between AI-assisted output and student-developed thinking.
Vermont has the better philosophy. China has the better implementation infrastructure. The question both are trying to answer — what kind of mind are we building, and what role does this technology play in that — is the one every teacher should be asking in every lesson plan, regardless of which document sits above it.
That question is ours to answer in the classroom, whatever the policy document says. And right now, that means we’re all deciding — through what we assign and what we require students to do first — whether the next generation is one that uses AI or one that thinks alongside it.
David Jacobson is a high school history teacher in Shanghai. He writes about AI and education at shouldiuse.ai and is the author of The AI Doesn’t Know Your Students.
