South Korea's $850 Million AI Textbook Bet Just Collapsed

South Korea spent roughly $850 million building a flagship program to put AI-powered textbooks in classrooms nationwide. Four months after the rollout began, lawmakers voted to strip the technology of its legal status as an official teaching material, effectively killing it. That expensive, very public experiment sits at the center of a debate that also played out, in miniature, on a recent episode of Students Incorporated, the podcast produced by students at the International Community School of Bangkok. Host Mr. Jason and student co-hosts James, Esther, Vijay, and Highlight spent an episode picking apart real-world case studies from the United Kingdom and South Korea to answer one question: does artificial intelligence make students sharper, or does it just make them dependent?

A Flagship Program That Lasted One Semester

The numbers behind South Korea's collapse are stark. According to The Korea Herald, the government allocated at least 533.3 billion won, about $385 million, to the AI textbook initiative in its most recent budget year alone, on top of roughly $850 million spent overall once equipment and teacher training are included. Piloted in the first semester of 2025 for English, math, and computer science classes from elementary through high school, the textbooks were supposed to personalize instruction for every student using adaptive AI. Instead, adoption stalled at around 30 percent of schools, and backlash from parents and teachers over the government's rushed rollout mounted quickly enough that the Ministry of Education abandoned its plan for a nationwide mandate in favor of school-by-school opt-in.

That retreat wasn't enough to save the program. In August 2025, South Korea's National Assembly passed a bill narrowing the legal definition of a textbook to printed books and e-books, cutting AI-powered materials out entirely. The reclassification stripped schools of the funding basis for their AI textbook subscriptions overnight. Publishers who had built entire product lines around the mandate, having sunk an estimated 800 billion won (about $580 million) into development, warned of layoffs and filed lawsuits against the Education Ministry; a 2026 report from the Seoul-based outlet Seoulz found that some are now pivoting toward Korea's private tutoring academies, while others are trying to export their AI textbook engines to friendlier markets in Southeast Asia.

The Same Debate, Argued by Teenagers

The Students Incorporated episode covered South Korea's reversal as fresh headline news, just weeks after the decision, and the tension its co-hosts explored is exactly what played out in Seoul. The show's UK case study centered on government-backed AI tools that grade student writing and free up teacher time, letting teachers "spend more time teaching and supporting students," as the segment put it. The upside, the co-hosts agreed, was real: less time buried in marking, more time on students who need it. The catch, one co-host said, was accuracy: "Imagine if an AI says your essay is average when you've actually written something creative that doesn't fit the AI's criteria." Another framed the trade-off bluntly: "It's efficient. But is it fair?"

The South Korea case study, drawn from the country's now-defunct push toward AI tutors that adapt lesson pace to individual students, drew a similarly split reaction. One co-host called personalized pacing "pretty cool," noting that with 30 students in a single classroom, a human teacher can't realistically give everyone that kind of individual attention. But another pushed back on what gets lost: "If you're always learning alone with AI, what happens to teamwork? Part of school is learning how to collaborate, discuss, and solve problems with other people." By the end of the segment, the group had landed on a shared verdict: AI in the UK model saves time but risks losing the human feedback loop that makes learning stick; AI in the South Korea model personalizes pace but risks isolating students from the social side of school.

I see it as a co-pilot. It handles the heavy lifting while we steer the direction.

That line, from one of the show's student co-hosts, doubles as a fair summary of where the discussion ended up: AI as an assistant, not a replacement, for the people actually doing the learning and teaching.

A Parallel Crisis: Cheating Outpaces the Tools Built to Catch It

In the episode's headline news segment, the co-hosts cited a Guardian report on the scale of AI-assisted cheating in UK universities, and the underlying numbers back up the alarm. A Guardian freedom-of-information investigation published in June 2025 found nearly 7,000 confirmed cases of AI-related academic misconduct in UK universities during the 2023-24 academic year, a rate of roughly 5.1 cases per 1,000 students, up from 1.6 per 1,000 the year before. A separate February 2025 survey by the Higher Education Policy Institute found that 88 percent of UK students were now using generative AI for assessments, up from 53 percent the previous year. Detection hasn't kept pace: more than a quarter of universities surveyed don't even track AI-related misconduct as its own category, and a widely cited study out of the University of Reading found that AI-generated exam answers submitted under a real student's name went undetected by markers 94 percent of the time, often scoring higher than genuine student work.

That gap between how fast students are adopting the tools and how slowly institutions are catching up is exactly the "efficient but is it fair" tension the ICS Bangkok co-hosts identified, just playing out at a much larger, messier scale than a single classroom.

Where the Money Is Actually Working

Not every AI-in-education story is a cautionary one. In Louisiana, a state-backed AI reading tutor called Amira, which listens to students read aloud and gives real-time coaching, has scaled from roughly 25,000 students in the 2023-24 school year to 100,000 students across 25 districts by the 2025-26 year. In December 2025, Louisiana's education department committed nearly $6 million in federal grant funding to expand Amira alongside two other AI-based platforms, on top of a separate $15 million federal literacy grant. Unlike South Korea's mandate-first approach, Louisiana's rollout has leaned on gradual, district-by-district adoption tied to measurable reading gains, the kind of incremental scaling that survived where a top-down national mandate did not.

The UK, meanwhile, course-corrected in the other direction the ICS Bangkok co-hosts worried about, not by pulling back from AI but by narrowing what it's trusted to do. In June 2025, England's Department for Education issued formal guidance permitting teachers to use AI for marking, but only for "low-stakes" work like quizzes and homework, and only with a teacher reviewing every output before it reaches a student. Early adopter schools in Liverpool and Kent reported saving up to six hours a week per teacher under the new guidance, without removing a human from the final decision.

What the Research Says the Students Already Sensed

The pattern connecting South Korea's collapse, Louisiana's growth, and the UK's cautious middle path shows up in the academic research too. A randomized controlled trial published in Nature's Scientific Reports in June 2025, led by Harvard researchers, found that a well-designed AI tutor outperformed in-class active learning by a substantial margin, with an effect size between 0.73 and 1.3 standard deviations. But a January 2026 Brookings Institution review of that study and others by education researcher Mary Burns drew a pointed conclusion: the programs that actually work don't replace teachers, they pair AI with them in what Burns calls "hybrid vigor." In the successful trials, human teachers still "play an essential role in monitoring and guiding students' use" of the technology, freeing them to spend classroom time on the discussion, critical thinking, and collaborative work that Burns's report says AI alone still can't deliver.

That's a research-backed version of the same conclusion the Students Incorporated co-hosts reached on their own, working through two case studies and a stack of quickfire questions from Mr. Jason. Used as a co-pilot, AI can speed up grading, personalize pace, and stretch a student's thinking. Used as a substitute for a teacher's judgment or a student's own effort, it just moves the same problems, unfair feedback, lost teamwork, eroded critical thinking, onto a bigger, more expensive stage. South Korea found that out the hard way, to the tune of $850 million. A handful of Bangkok teenagers got there in about twenty minutes of conversation, no failed national rollout required.

← Back to all posts