Measuring the real impact of AI in education

What does it actually take to know if an AI tutor is helping kids learn? Bibi Groot, Chief Impact Officer at Eedi Labs, has spent her career answering exactly that question — first at the Behavioral Insights Team (aka the Nudge Unit, co-founded with Nobel laureate Richard Thaler), then in classrooms across the UK and Latin America.In this episode, Bibi walks us through how Eedi's diagnostic engine works — 60,000 carefully designed multiple-choice questions, each distractor linked to a specific misconception — and why understanding why a student gets something wrong matters as much as knowing they got it wrong.Bibi also introduces a concept that should alarm everyone in edtech: cognitive surrender — the risk that when AI does all the thinking, students stop learning altogether. Her solution is architectural: don't ask students to self-regulate, build the constraints directly into the system. She references a striking study by Poulidis and Bastani on chess students — those who received AI hints at system-chosen moments improved 64% vs. only 30% for those who could ask for help whenever they wanted.This is a rare, rigorously evidence-based conversation about what responsible AI tutoring actually looks like — and how far most of the field still has to go.References mentioned in this episode:Behavioral Insights Team (the Nudge Unit)Eedi Labs — including the free Eedi School platformGoogle DeepMind's LearnLMLearning Engineering Virtual Institute (LEVI) — created by Schmidt Futures & Renaissance PhilanthropyDavid Yeager, 10 to 25: The Science of Motivating Young PeopleAI Hub for Education (Stanford) — reviewed 800+ papers on AI in education; only 20 had causal evidencePoulidis & Bastani chess study (system-chosen AI hints → 64% improvement vs. 30% for on-demand help)London EdTech Week — Meet Bibi & Svenia at the London AI & Education Meetup on June 18, 2026