Students get AI boost in writing, blow to recall

AI helped college students write better essays, but it cost them comprehension, an experiment by the JoongAng Ilbo found.

Students from Seoul universities gathered in Mapo District, western Seoul, on Sept. 9 and 10, where they were given 30 minutes to write an essay on what should take priority in a democratic society when expert judgment conflicts with the will of the majority. One group used AI, while the other did not. A five-member panel of experts, including professors of Korean literature and education, evaluated the essays. 

The AI group scored an average of 81.6 points, well above the 68.8 points scored by the group that didn't use AI.

The biggest gap was in expression, including sentence construction. The AI group scored 82.7 points on appropriateness of expression, more than 20 points above the 61.3 points of the group that didn't use AI.

"The somewhat clumsy, unfinished pieces made me think a person had written them, while the polished ones, with no problems in logical structure and no spelling mistakes, made me think AI might have written them, and the scores came out as I expected," said Noh Dae-won, a professor of Korean language and literature at Sungkyunkwan University who sat on the panel.

The picture flipped when participants were tested on what they had written. Right after the essay, each was given follow-up tasks, such as rewriting a key sentence and listing the evidence supporting their argument in order. The AI group's average comprehension score was 66.9 points, below the 82.9 points of the group that didn't use AI.

Every participant who wrote without AI could recall the final sentence of their essay, while more than half of those who used AI could not. The AI group also scored lower on recalling evidence and supporting details and on understanding the logical structure of their paragraphs. While AI users generally remembered their central argument, they recalled relatively little of the evidence and specific information used to support it.

"Honestly, there was not a single sentence I could clearly rewrite," said Kim Seo-eun, aged 23, a Konkuk University student in the AI group. "I was surprised at how much I realized this wasn't my writing."

Hwang Hyeon-min, aged 24, a Seoul National University student, said he was left with regret. "I feel that if I had written it myself, I could have remembered it better," he said.

Yoo Seong-ho, a professor of Korean language and literature at Hanyang University, said the results suggest a lack of ownership. "If there are many parts you cannot put into words yourself, it means you haven't internalized or taken ownership of what you learned," he said.

Han So-won, a professor of psychology at Seoul National University, said the boundary between a person's ability and a tool's ability blurs once the thinking itself is handed off. "Once you start handing the work of thinking over to a tool like AI, the line between where your ability ends and the tool's ability begins gets blurry," she said after seeing the results. "If you study by relying on AI that gives quick answers, you may lose the ability to think things through yourself."

The findings come as reliance on AI grows. A 22-year-old student at a private university in Seoul, identified only by her surname Kim, alternates between ChatGPT, Claude and other generative AI models dozens of times a day. She uses them to summarize recorded lectures and find materials, and she says AI is essential for writing assignments and preparing for exams.

"I ask each AI [model] to evaluate the answers other [models] give, then pick the one that looks best to solve the problem," she said. "Sometimes I feel small, thinking all I do is connect the AI [models] so they can talk to each other, but I can't imagine living without AI."

Some scholars have warned that thoughtlessly outsourcing thinking to AI could produce the first "generation of intellectual decline" in human history.

Concern is growing overseas as well. Jared Cooney Horvath, a U.S. neuroscientist, said in a report submitted to the U.S. Senate in January that Gen Z, defined as those born between 1997 and 2010, is "the first generation in modern history to score lower on standardized academic tests than the one before it."

He said the Flynn effect, in which IQ had risen with each generation since intelligence measurement began in the late 19th century, has stopped and turned downward for the first time. 

The Organisation for Economic Cooperation and Development's (OECD) Programme for International Student Assessment 2025 report, released on Sept. 8, also showed that the average academic achievement of students in 23 major OECD countries last year fell from the previous survey in 2022.

BY OH SAM-GWON, CHOI HYE-RI [[email protected]]