Last Thursday I had the opportunity to attend the presentation of the PISA results, the same ones that today dominate headlines in Spain and around the world, at the OECD offices in Paris. It was an extremely interesting presentation, and at the end of the briefing I was very clear about the question I wanted to ask the Director of Education: “So, regarding AI, what would be the main message for the policymakers?” I asked. His answer was simple: “We should be careful… very careful”.
And no, it wasn’t a catastrophist and hurried warning, but a response dictated by prudence — I would dare to call it common sense — in light of the clues left by the results of PISA 2025. The main one? Students who report using AI daily systematically obtain worse results, between twenty-three and twenty-six points lower than those who say they use it little or never… but there is more.
“Students who report using AI daily obtain, in a systematic way, worse results”
In the 2025 edition of PISA, for the first time, the students from 91 countries and economies around the world were asked about the frequency and the type of use they give to AI to complete their homework at home. The results, although preliminary, are suggestive. When students use AI to replace cognitive processes of reasoning and creation, such as drafting an initial text, summarizing readings, or conducting research on a topic of interest, then that use is negatively correlated with the science score, which falls by nineteen to twenty-nine points compared to those who never use it (even after controlling for socioeconomic status). This correlation softens and disappears only for students who report using AI as study support.
Average science score by frequency of AI use for school tasks
Average science score by category.
Unadjusted Adjusted for socioeconomic status
And yes, correlation does not imply causation. The usage data reported by the students themselves can carry all sorts of biases, but none of those cautions authorizes us to look the other way. We do not have five years to wait for unequivocal causal studies; we need to decide now, today, the path that educational policy and classroom pedagogy will take regarding AI. The minimum we should ask for is prudence, take a healthy distance from a rushed and unrestricted adoption of AI in educational centers and undertake the challenge of stimulating creative processes beyond mere repetition. If nothing changes, students who seek the easy path will always be able to “delegate” their tasks to the AI of the day after school every afternoon.
“When students use AI to replace cognitive processes of reasoning and creation […] that use correlates negatively with the science score”
The drop in scores in 2025 is substantial and places performance at the lowest point since PISA first took place in 2000. In the last decade, the OECD’s average score has fallen by twenty-eight points in reading, twenty-two points in mathematics, and seven points in science. Of course, it would be naive to attribute the sharp decline solely to the emergence of AI, just as it would be naive to try to exhaust all the factors that could explain it in this article. Instead, let’s focus on one of them: mobile devices and the use of social networks.
The results of PISA show a worrying pattern: schools devote more hours to device use than to recess, between two and three hours of the school day. And, just as with AI, when analyzing students’ responses we find another suggestive correlation: among those who limit themselves to one hour daily and those who exceed five, there is almost eighty points difference in science, and more than one in four students acknowledges that their classmates are distracted by mobile phones in most classes. Unknowingly, by introducing mobile devices both in the classroom and in leisure time, the screen gradually absorbs those dead moments, the moments of boredom, of imagining… those moments in which ideas settle and we allow ourselves to create.
Ya lo decía Jonathan Haidt en su famoso libro The Anxious Generation (2024). Between 2010 and 2015 there occurred what he calls the great reconfiguration of childhood, the moment when play-based childhood gave way to phone-based childhood, with sleep deprivation, social isolation, fragmentation of attention, and addiction as consequences. His thesis has sparked controversy in the academic world and not all researchers share the force with which he attributes the adolescent mental health crisis to screens, but it is a fact that the data he laid on the table years ago was pointing in the same direction as what PISA points today.
What should we do about all this? Digital innovations advance at breakneck speed and AI, like screens before it, has come to stay. And yes, they open enormous possibilities for growth and development. But there is a stage that does not admit shortcuts: a child, a teenager learning to think for themselves, to analyze, to hold a personal idea and to defend it. Let us protect that moment! Let us care for human intelligence to unleash the maximum potential of the artificial. And may that reminder linger in the background, as a reminder, that answer I received in Paris: we must be careful… very careful.
