🔴 Must Know
A randomised trial finds teacher AI tools lowered student motivation — and hurt weaker teachers’ classes. Alp Süngü (Wharton), Benjamin Lira Luttges and Angela Duckworth ran a randomised experiment across 14 schools in Turkey — 193 teachers, 2,816 students, 14,198 course observations. Teachers given a GPT-4o-powered planning and assessment assistant produced no average gain in student achievement, and their students rated courses as less interesting, less enjoyable and less important (intrinsic motivation down 0.11 SD). The null on achievement hid a split: students of lower-performing teachers scored 0.129 SD worse. Usage logs explain a lot — 66% of AI conversations were material generation, and the median conversation was two prompts. This is substitution of pedagogical effort, not augmentation. It is the most rigorous evidence yet that “AI saves teachers time, therefore learning improves” is an assumption, not a finding — and it lands squarely on the most common way schools deploy AI today.
EU AI Act enforcement begins 2 August — three weeks away. National market surveillance authorities start supervising and enforcing Article 4 (AI literacy) from 2 August 2026, and most remaining obligations, including the high-risk requirements that cover AI used in admissions, assessment and monitoring, apply from the same date. Any school operating in the EU — including international schools with EU campuses — is a “deployer” and must be able to evidence sufficient AI literacy among staff. Schools outside the EU should watch: Article 4 is becoming the reference standard other regulators copy.
Subscribe to continue reading
Subscribe to get access to the rest of this post and other subscriber-only content.