A third-year high school student studies independently at Sungji Girls’ High School in Masanhappo-gu, Changwon, Gyeongsangnam-do. Yonhap News
The International Artificial Intelligence Ethics Association called on education authorities to reconsider from scratch their plan to use artificial intelligence to grade written and essay questions on South Korea’s college entrance exam.
In a statement on the 27th, the association said it does not oppose expanding written and essay assessments themselves. But it said the existence of a technical means to clear grading bottlenecks and the acceptability of using that means in college admissions, which shape a person’s life, are entirely different questions.
The National Education Commission and the Korea Institute for Curriculum and Evaluation had each proposed, as part of a broader overhaul of the college admissions system, having AI grade written and essay answers on the College Scholastic Ability Test first, with teachers making the final determination. The commission plans to gather public opinion through a deliberation process and present its admissions overhaul plan in late October.
The association argues that entrusting exam grading to an incomplete technology, one in which bias in training data and hallucination — the fabrication of content that is not factual — remain unresolved, is highly risky. It was particularly skeptical about the approach of having human evaluators provide final verification.
“Given the reality that hundreds of thousands of answers must be reviewed within a limited period, automation bias, in which teachers uncritically accept AI’s initial drafts, could occur,” the association said. It also criticized the plan, saying it is difficult to clearly explain to students the valid grounds for scores assigned by AI, and that there is no procedure at all to provide redress if an error surfaces belatedly after admissions results have been finalized on the basis of faulty grading.
Citing the AI Framework Act, which took effect this year, the association said that because exam grading goes beyond simply evaluating students and determines the significant individual right of university admission, it falls under the “high-impact AI” category specified in the law. That means strict guidelines, including transparency of criteria and re-verification procedures, must be established before AI grading is introduced.
“The fact that AI grading has become technically feasible does not immediately mean it may be used in college admissions,” said Jeon Chang-bae, chairman of the International Artificial Intelligence Ethics Association. “Because college admissions are directly tied to the lives of all citizens, the decision must be made cautiously, through in-depth student- and human-centered research and social consensus, so that not even a single student suffers unfair harm.”
