Korea must rethink math education in the AI era

Columns With artificial intelligence acing olympiad problems, Korea must shift its math education from a focus on speed and correct answers to reasoning, judgment and student interest. Front row, from left: 2026 Fields Medalist winners Chinese mathematician Yu Deng, U.S. mathematician John Pardon, Canadian mathematician Jacob Tsimerman and Chinese mathematician Hong Wang pose for photos onstage after receiving their Fields Medals during the International Congress of Mathematicians in Philadelphia on July 23. AFP/YONHAP Kwon Oh-nam The author is the president of the Korean Federation of Science and Technology Societies and a professor of mathematics education at Seoul National University. The Fields Medal, awarded every four years at the International Congress of Mathematicians, is often called the Nobel Prize of mathematics. Four winners were recently announced at the congress in Philadelphia. Among them were Hong Wang of New York University and Yu Deng of the University of Chicago, Chinese nationals who solved problems that had remained open for more than a century. As a mathematician, I ask a question as the world focuses on mathematics: What have we been teaching as mathematics in school? The question is urgent. Before the Fields Medals were awarded, six AI systems each solved all six problems at the International Mathematical Olympiad in Shanghai for perfect scores. They did so within the time limit without human intervention. Of 666 students who tackled the same problems, only seven earned perfect scores. If AI can do this, what have we been teaching? Largely, the ability to reach a predetermined answer within a fixed time. Schools evaluate students accordingly, competitions identify talent this way and college admissions use mathematics scores to divide students among academic paths. Korea's College Scholastic Ability Test gives students 100 minutes to answer 30 mathematics questions — three minutes and 20 seconds each. We have long said the process matters more than the answer. But that is no longer enough because AI can provide solutions and proofs. We should demand three things: First, students should choose what problem to solve and how, then explain why. Even with the correct answer, they do not truly understand if they cannot explain why they chose a particular route. Second, students should compare alternative solutions, solving a problem two ways and explaining why one is better. Third, they should find flaws in a proof placed before them. AI answers can look formally sound while being plausibly wrong. If students cannot spot a missing step, the error passes unnoticed. None of these abilities is currently assessed in our classrooms. Whenever I present Korea's case at international mathematics conferences, I hear the same question: Why are Korean students among the world's highest achievers yet near the bottom in interest? Students do not value abilities that are not assessed. As long as evaluation measures only correct answers and speed, mathematics becomes a subject to endure. The three abilities above cannot be developed simply through instruction. They require genuine interest. The International Association for the Evaluation of Educational Achievement surveys fourth and eighth graders' mathematics and science performance every four years through the Trends in International Mathematics and Science Study. In 2023, Korea ranked third worldwide at both grade levels. Yet in Korea's 2025 National Assessment of Educational Achievement, only 40.7 percent of ninth graders said they were interested in mathematics. This is not merely an emotional issue. Mathematical self-efficacy strongly predicts achievement, and causality runs both ways. Students headed toward science and engineering tend to have greater self-efficacy. Falling confidence among ninth graders can foreshadow lower applications to science and engineering programs three years later. Interest is not an end in itself but the soil in which mathematical ability grows. Three things must change: assessment, classroom instruction and the questions used for both. First, assessments should test judgment and explanation. Concerns about objective grading are legitimate. In public schools offering the International Baccalaureate, external mathematics assessments grade reasoning and justification, not merely whether an answer is correct. Teachers practice grading actual student responses and receive training to standardize scoring. Second, classrooms need lessons in which students examine and challenge AI-generated proofs. Rather than banning AI, schools should teach students to judge its answers. Teachers must first be given the time and materials to design such lessons. Third, we need a growing bank of questions asking students to evaluate AI answers. University researchers in mathematics and mathematics education should work with classroom teachers to develop questions and scoring criteria. Some tasks should also ask students to formulate problems themselves. The National Education Commission is discussing whether to add constructed response and essay questions to the CSAT. Whatever it decides, Korea needs suitable questions and scoring standards. There is no reason to wait for a draft proposal expected in late October. This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.