AI-powered textbooks fail to make the grade in South Korea (restofworld.org)

South Korea’s high-profile experiment with “AI-powered” digital textbooks—deployed in March for math, English and computer science as a flagship education program—was quietly stripped of official textbook status after one semester. Students and teachers reported widespread technical glitches, factual errors, higher teacher workload, increased screen time and unclear personalization. Adoption plunged from 37% in the first semester to 19% for the new term; the government spent ~1.2 trillion won ($850m) and publishers invested ~800 billion won ($567m). Critics also flagged data‑privacy risks and a rushed development cycle (AI titles were produced in roughly 12/3/3 months for development/review/prep versus the typical 18/9/6), prompting lawsuits, political pushback and the program’s reclassification as supplementary material. The episode is a cautionary case for the AI/ML community: embedding adaptive systems in classrooms requires careful piloting, robust content verification, explicit data governance and curricular integration rather than top‑down mandates. Developers say they followed security rules and didn’t store personal data; proponents note potential benefits for underserved or disabled students. But the rapid rollout, mixed pedagogical outcomes and political fallout underscore practical risks—model accuracy, UX for young learners, teacher workflows and trust—anyone building or deploying AI learning tools must address before scaling.