800 studies, 20 that prove anything

Stanford's SCALE team reviewed the research on AI in schools this year.

Out of more than 800 papers they found 20 with strong causal evidence, and not one high-quality causal study of a student-facing AI tool in an American K-12 classroom. The pattern in the studies that do exist is the interesting bit. Kids perform better on maths, coding and writing while the tool is in front of them, and when you take it away and test them on their own the effects are mixed. Performance goes up, learning maybe not. B

rookings went further in January and said the risks to children currently outweigh the benefits.

I'm hardly neutral here, I'm the one proposing an AI-native primary school, and the honest position is that the evidence for my thing doesn't exist yet either. So when a vendor turns up at your school with a graph, ask whether the effect was measured with the tool switched on or switched off.