Let pupils build and question a model
Can building a simple model help children understand AI? Across 11 Madrid-area schools, 996 pupils aged 10-11 completed four sessions using visual programming and tools such as Teachable Machine. They created and trained models rather than only consuming AI-generated answers. Average scores on a ten-point test rose from 5.54 to 6.24; the test covered machine-learning foundations and supporting programming concepts. Nearly 95% reported enjoying model creation. This is encouraging evidence that an accessible, practical unit can accompany improved knowledge and strong interest. It cannot show that the unit caused the gains: there was no comparison group, and the analysis did not account for pupils sharing classrooms and schools.
Why it matters for leaders
Pilot a short build-test-explain unit. Ask pupils to change training examples, find a failure and explain why it occurred. Include an unfamiliar classification task afterward, with individual explanations and accessible alternatives for pupils who need them.
The evidence
Purposive sample from 17 classrooms; paired pre/post tests reported a standardized change of about 0.51. The article identifies this dataset as distinct from earlier related work.
Who was studied
Primary pupils in Spain; direct K-12 evidence.
What the evidence cannot tell us
Immediate results, unmeasured prior programming experience and clustered pupils limit inference. Enjoyment and knowledge scores do not establish broader critical-thinking transfer.
Read the original sources
Summary and leadership analysis from the Week ending 2026-10-23 brief. The edition’s selection notes explain source access and evidence limitations.