MIT has just published a guide on AI in schools, A Guide to AI in Schools: Perspectives for the Perplexed. Thanks to Remco Pijpers for pointing me to it. Right in the introduction, Justin Reich writes that a guide to AI in 2025 is akin to a manual for aviation in 1905: just after the first flight, but long before anything resembling a flight plan existed. It’s an apt comparison. Everyone senses that something is taking off, but no one really knows where it’s heading. We don’t yet know which knots to tie, or how tight or loose they should be, but we’re learning as we go.

To be clear, the guide wasn’t written to provide answers. It was written to raise the right questions. What does it mean, for instance, when a machine does your homework? Or when teachers let a language model generate their feedback? And how fair is it to ban AI tools when some students can use other devices at home?

What makes this MIT guide refreshing is that it begins in classrooms, not Silicon Valley. A hundred teachers, students and school leaders were interviewed, and their voices form the heart of the document. They sound strikingly human: curious, worried, sometimes openly sceptical.

There’s no plea for or against. The tone is more like: try something, evaluate it, and change it if it doesn’t work. As Reich notes, an AI policy should never be a “forever policy”. Better to call it your 2025–2026 policy. It sounds temporary. And that’s exactly the point.

The guide covers big themes from ethics and privacy to motivation and attention. At the same time, it resists the usual urge of policy papers to regulate everything. It reads more like a staffroom conversation than a list of directives. A conversation about how to deal with new technology while you’re still in the middle of the school year, juggling exams, reports, and pupils who mostly want to know whether their work is any good.

What stays with you is that technology rarely solves problems by itself. It can save time or waste it, personalise or standardise, inspire or dull. And perhaps that’s the main lesson: AI is not the pilot, merely a passenger who occasionally grabs the controls uninvited.

What Schools Can Actually Do

The guide also offers a set of simple yet thoughtful suggestions for schools that want to start working with AI but are unsure of where to begin.

Start small, and make it temporary.
As I mentioned earlier, call your AI guidelines your 2025–2026 policy. It may sound trivial, but it creates space for learning. Policy becomes an experiment, not a law.

Include the sceptics.
A good AI team doesn’t consist only of enthusiasts. The critical voices help you identify pitfalls that others may miss.

Begin with values, not tools.
Talk first about what your school stands for — honesty, creativity, autonomy, privacy — and only then about which applications fit those values.

Let teachers experiment first.
Encourage staff to use AI for creating rubrics, letters, or lesson ideas before introducing it into the classroom. It’s safer — and more instructive.

Build AI literacy, including among leaders.
Anyone developing policy without understanding how AI works is unlikely to create sound policy.

Avoid blanket bans.
They rarely work and often widen inequality. Students with more resources will find their way around them anyway.

Use real cases to discuss boundaries.
Like the teacher who used AI to create action figures of pupils. Creative, yes, but full of privacy dilemmas. Examples like these bring the debate to life.

Treat AI as practice material, not as a finished product.
Let students analyse, critique and improve AI-generated texts. It’s the modern version of learning to write in the margins.

Above all, keep listening.
The best insights don’t come from universities or tech firms, but from classrooms where teachers, every day, try to fly the plane while still building it.

The guide resonates strongly with what I discussed in my recent talks at ResearchED Cambridge and Edinburgh: that the challenge is not just to use AI wisely, but to stay human while doing so.
You can read the complete guide here: tsl.mit.edu/ai-guidebook.