Post by AI World

Speed, accuracy, and completion rates are the metrics being used to evaluate AI in schools. None of them measures learning. The OECD - OCDE's 2026 Digital Education Outlook shines a light on what is happening when students use AI in the classroom. The cognitive effort is outsourced, and what Dragan Gasevic calls metacognitive laziness takes its place. While AI use improves everyone's scores, once access is removed students perform worse than peers who never had it. In one cited study, by 17% on maths tasks. Four effective uses of GenAI are mapped in the report:  ➜ As a tutor: scaling personalised learning, including in low-infrastructure settings with offline small models ➜ As a partner: collaborative learning, with students and teachers using GenAI together inside instruction ➜ As an assistant: lesson planning, feedback, administrative load, freeing teacher judgement ➜ At system level: research, learning pathways, study advisors, credit mobility across institutions What is harder to design for is the second digital divide the OECD report surfaces. Students in well-supported settings learn to use GenAI as a coach. Students in less-supported ones use it as a shortcut. The same tool, deployed without pedagogical design, widens the gap it was meant to close. The OECD has set out the design criteria that separate real learning from metrics performance. We at AI World track how the ecosystem is developing. Credit to the CERI team and contributors for resisting easy narratives: Andreas SCHLEICHER, Stéphan Vincent-Lancrin, Ryan Baker, Dragan Gasevic, Ronald Beghetto, Seiji Isotani. #AIandEducation #GenAI #EdTech #OECD