Abstract:
OECD's latest PISA data found that students who use AI-assisted learning tend to perform worse overall than students who do not use AI, but if AI use is combined with training on "how to evaluate AI results," the situation will be different.
The OECD study titled PISA is based on data collected in 2025 and is the first survey of its kind since AI has truly become popular. The study covered more than 760,000 15-year-old students in 91 countries and regions, and tested subjects including science, mathematics and reading.
Research shows that after controlling for socioeconomic background, students who never use AI generally outperform students who use AI in science. However, the gap is not entirely consistent: students who use AI only to summarize texts they are supposed to read, or to draft writing assignments, tend to perform worse; by comparison, students who use AI for preliminary research, or more generally to "help me study," see smaller declines.
The frequency of AI use will also affect the results. Research has found that students who use AI only once or twice a year, and students who use AI every day, tend to perform the worst; students who use AI every month or every week have relatively better results.
But AI is not completely without positive effects. Research points out that students who use AI every week to generally assist learning have slightly higher grades than students who do not use AI; and if students also receive regular training on how to judge the quality of information generated by AI, this advantage will be further expanded. The study authors therefore believe that moderate and intentional use of AI in schoolwork and learning may bring positive results.
Andreas Schleicher, director of learning and skills at the OECD, stated in the article that learning does not occur through passive reception of content like "watching sports games", but comes from an active cognitive struggle between the brain and new materials; when technology can enhance this struggle, students will progress, and conversely it will weaken students' development.
The study also found that AI use is more common among students from relatively advantaged backgrounds, which may be related to their easier access to hardware, stable networks, and paid AI tools. There are also wide differences between countries: more than 95% of students in Vietnam reported using AI tools, compared with about 60% in Japan.
In addition, AI users are generally more curious, with students who use AI every day having the highest curiosity levels. However, at least for now, this curiosity has not steadily translated into better results.
Comments