AI에게 요약·초안작성 시킨 학생, 그렇지 않은 학생보다 성적 낮아

A study of 82 countries revealed that students who assigned artificial intelligence (AI) to summarize or draft essays had lower academic performance compared to those who did not. However, students who only used AI to ask questions performed better than those who didn't use AI at all, suggesting that the way AI is used significantly impacts academic achievement.

The OECD's 2025 Programme for International Student Assessment (PISA) surveyed students in these countries about their use of AI for schoolwork. Specifically, it asked about four uses: summarizing assigned reading, drafting essay outlines, preliminary research on new topics, and providing learning assistance. Korean students reported using AI for these purposes more frequently than the OECD average, with "learning assistance" at 50.7% (compared to the average of 45%), "drafting essay outlines" at 37.4% (28.8% average), "preliminary research on new topics" at 43.9% (31.2% average), and "summarizing assigned reading" at 36.8% (29.7% average).

When students used AI for summarizing or outlining assignments, their scores were lower than those who didn't. For instance, students who asked AI to summarize assigned reading scored 31 points lower (477 points vs. 508), those who used AI to draft outlines scored 29 points lower (481 vs. 509), and those who used AI for preliminary research scored 19 points lower (486 vs. 505).

Among AI users, those who used AI 1-2 times a week performed better than those who used it less than once a month or nearly every day. However, using AI for learning assistance did not always lead to lower scores. Students who used AI to seek help during learning difficulties scored only slightly higher (501 points) than those who didn't use AI at all (499).

The OECD notes that Korean students were ahead on average when it comes to AI literacy education. 66.5% of Korean students reported having received classes on evaluating the quality of information generated by AI in school, compared to the OECD average of 62.6%. These students also tended to have slightly higher achievement scores even if they used AI frequently.

However, OECD warns that students from disadvantaged socio-economic backgrounds may not have equal access to such education, highlighting a new form of educational disparity in the AI age.