Jung Min-ho has worked as a staff writer at The Korea Times since 2012, mostly covering social and political issues. He currently belongs to the Politics & City Desk where he covers topics such as health, labor and human rights. Prior to joining the team, he was responsible for covering North Korea and sports. His article about a biosecurity breach of Middle East respiratory syndrome won him an award from the Korea Science Journalists Association in 2016. He is also the co-author of the book, "Medical Pioneers of Korea" (2019). He served as the head of the international relations committee at the Journalists Association of Korea from 2021 to 2023.
Korea’s infamously difficult CSAT English section to get AI makeover
Students prepare for the College Scholastic Ability Test at Hyowon High School in Suwon, Gyeonggi Province, Nov. 13, 2025. Newsis
By Jung Min-ho
AI system to help write, review English passages after uproar over exam’s difficulty
The notoriously difficult English section of Korea’s College Scholastic Ability Test (CSAT) is now heading for an artificial intelligence (AI)-driven overhaul, with officials saying that the reform begins with a simple diagnosis: The English used on the university entrance test is not the kind of English people actually use.
The Ministry of Education announced earlier this week that it will build an AI-based system to generate English passages, with the goal of using them in mock exams for the 2028 academic year. Officials will also work to apply the technology to review CSAT questions.
A senior ministry official who oversees the project acknowledged that recent CSATs included English passages using language that is “not actually used in real life.”
“The reason is that we have been relying on overseas source texts, and those sources tend to be academic papers or convoluted scholarly books, which have drawn criticism (for being impractical),” the official told The Korea Times.
He said the problem of CSAT English being impractical and unnatural is part of what the ministry is trying to fix through the ongoing work for the new AI system.
Once the system is built, AI is expected to generate and revise passages “in interaction with” human experts, according to the official.
The AI initiative comes as a reminder of what CSAT English is officially for. According to the Korea Institute for Curriculum and Evaluation, which designs and administers the CSAT and other national-level educational assessments, the purpose of the 2025 CSAT English was to assess how well students have met high school English curriculum standards, as well as their “practical English skills for everyday life” and the English proficiency needed for college study.
That means, in principle, the exam should sit somewhere between everyday communication and academic reading. But critics say recent CSATs have substantially missed that target by heavily favoring complicated sentences and pseudo‑academic structures that few encounter outside of test papers.
For some experts, the shift to AI is an overdue chance to bring the exam closer to its stated purpose.
Because modern language models are trained on massive amounts of real‑world text, Jeong Han‑min, an AI expert at the Korea Institute of Science and Technology Information, expects AI‑assisted questions to resemble authentic English more than the “twisted prose” that has become infamous in recent exams.
Even if humans develop most test questions, AI could act as a second reader, flagging overly long or convoluted sentences, evaluating the logical flow and offering quantitative insight into each passage’s difficulty, clarity and style, he told The Korea Times.
But Kim Kyeong-seo, CEO at VAIV Company, a Seoul-based firm specializing in AI, cautioned that AI alone does not guarantee more practical language. What the system produces will depend on the data it is trained on.
Kim said the style of future CSAT passages will reflect how far policymakers are willing to go to enforce the official mission of testing practical English rather than puzzle‑solving.
"Because large language models can be fine‑tuned exactly as you want, the outcome really depends on the intention behind that fine‑tuning," he said.
He sees particular potential on the evaluation side. Once draft CSAT passages are prepared — whether written by humans, AI or both — the system can estimate their difficulty and gauge how “practical” or “academic” the language feels. That kind of quantitative analysis is something no human review committee can easily match, he said.
Some educators remain uneasy. They worry that once AI is allowed to help create and review the CSAT English questions, it will be easier for it to spread to other subjects eventually, gradually eroding human authority over content.
“It’s very convenient to use AI for translation and other tasks, but I still believe humans must retain authority [over important exams], and I don’t think using AI to create test questions is appropriate,” said Moon Hyung-nam, professor at Sookmyung Women’s University in Seoul and chairman of the Korea Association of AI Education.