Students who use artificial intelligence for schoolwork generally perform worse than those who do not, according to new OECD education data, although the findings suggest that carefully managed use can improve learning.
The analysis comes from the latest Programme for International Student Assessment (PISA), which examined the experiences and performance of more than 760,000 15-year-olds across 91 countries. The assessment covers science, mathematics and reading, and is the first conducted since generative AI became widely used by pupils.
After taking students’ socio-economic backgrounds into account, those who said they never used AI typically achieved higher science scores than those who did. However, the size of the gap varied according to how the technology was used.
Pupils who relied on AI to summarise texts they were expected to read or to draft written assignments tended to show the weakest results. Those using it for initial research, or more generally to support their learning, experienced a smaller disadvantage.
Frequency also appeared to matter. Students using AI once or twice a year, as well as those using it every day, tended to perform less well than pupils using the tools weekly or monthly.
When AI use can support learning
The OECD found that weekly users who employed AI to help them learn generally achieved slightly higher results than non-users. The advantage was more pronounced among pupils who were regularly asked to assess the accuracy and quality of AI-generated information.
When that kind of critical evaluation was part of their education, regular users of AI for general learning support outperformed students who did not use the tools. The report’s authors said the findings pointed to possible benefits from “moderate and intentional use of AI for schoolwork and learning”.
Andreas Schleicher, the OECD’s director of learning and skills, said technology helped when it strengthened the mental effort involved in understanding new material, but harmed progress when it removed that effort.
“In the same way that we do not become fit by watching sports but by doing sports, learning does not occur through the consumption of content, but as a productive cognitive struggle of the mind with new material,” he wrote.
“Where technology enables or enhances that cognitive struggle, students will advance. Where technology short-circuits the productive struggle of learning, it will undercut students’ development.”
The data also showed that access to AI is uneven. Use was more common among students from relatively advantaged backgrounds, who are more likely to have suitable devices, reliable internet connections and access to paid tools.
There were substantial differences between countries. More than 95 per cent of students in Vietnam reported using AI tools, compared with about 60 per cent in Japan.
Curiosity was also associated with AI use and was highest among pupils who used the technology daily. But the report suggests that greater interest in AI has not consistently translated into stronger academic performance.
