Korea took AI textbooks national before the argument finished
The AI digital textbook programme reached classrooms in March 2025 as the world's first national rollout of adaptive learning software, with its legal status, its evidence base and its teacher training all still unsettled.
No education ministry had previously attempted what Korea’s did in March 2025: the simultaneous national introduction of adaptive learning software as an official textbook, beginning with mathematics, English and information studies in the third and fourth years of elementary school and the first years of middle and high school. The design of the AI digital textbook, announced in 2023 and developed by commercial publishers against government specifications, was straightforward in principle. Students work through material on a tablet; the system tracks response patterns and adjusts difficulty and sequencing; teachers see a dashboard of who is stuck where.
The promise is the oldest one in education technology and the hardest to dismiss. Korean classrooms run at roughly twenty-plus students to a teacher with a national curriculum and a fixed pace, which guarantees that some students are being taught material they already know while others fell behind two units ago. Software that meets each of them at their own level addresses a real structural failure rather than an invented one. Mathematics was chosen first for exactly this reason: it is the subject where accumulated gaps compound fastest and where diagnostic questions map most cleanly onto a skill tree.
The objections that accumulated through 2024 were of three kinds, and they were not interchangeable. The first was screen exposure — a country whose paediatric and psychiatric bodies had spent a decade warning about adolescent device use was proposing to make devices the medium of instruction, and parent petitions gathered signatures on that basis alone. The second was data: adaptive systems work by retaining fine-grained records of what each child got wrong, held by private publishers under contracts most parents had not read. The third was evidentiary. Personalized-learning trials internationally have produced mixed and often small effects, and Korea was scaling nationally without a domestic trial of comparable size.
The legal fight was separate from all of these and turned on a single word. Under Korean law a designated textbook must be adopted by schools; an educational material may be adopted at a school’s discretion. In late 2024 the National Assembly amended the Elementary and Secondary Education Act to reclassify AI digital textbooks as materials rather than textbooks, the government objected, and the status remained contested well into 2025. The practical outcome was that adoption became substantially a school-level choice in the first year, which is close to what a pilot would have looked like, arrived at by legislative dispute rather than by design.
The preparation gap was the quieter problem. The ministry ran training for large numbers of teachers ahead of the rollout, but the training was mostly about operating the platform, and the harder shift is pedagogical: a teacher whose class is working at fifteen different points in the curriculum is running a different kind of lesson, and no dashboard supplies the classroom management for it. Metropolitan and provincial education offices also absorbed the device costs, which fell unevenly across jurisdictions with different budgets and different existing stocks of tablets.
What Korea bought with the speed was information no other system has. A national deployment generates evidence at a scale that pilots cannot, and if the ministry publishes learning outcomes honestly against comparable non-adopting schools, the finding will be worth more than the programme. What it risked is the standard cost of moving first: a cohort of students used as the trial, and a technology whose reputation is set by its worst implementations.
The underlying judgement is defensible and was made in the wrong order. Korea has the network, the devices and the administrative capacity to do this before anyone else, and adaptive software probably does help students who are behind. Establishing that before mandating it nationally, and settling what the thing legally was before publishers had built it, would have cost two years and very little else.