Stop turning schoolchildren into crash-test dummies
AI can help in classrooms but it needs to be treated with caution
EVERY three years, hundreds of thousands of 15-year-olds sit down to answer questions in the global Programme for International Student Assessment (Pisa) test, now held in more than 90 countries.
They do not revise for it, nor do the results affect their grades. Yet, ministers scrutinise the outcomes, newspapers splash them across their front pages and education systems are praised or questioned accordingly.
This is because those outcomes are the most reliable measure of the quality of school systems, and thus one of the most powerful predictors for the long-run success of economies.
Researchers at Stanford University estimated that if all 15-year-old Americans achieved at least the Pisa baseline Level 2, this could add more than US$40 trillion – one-and-a-quarter times America’s nominal gross domestic product – to the country’s economy over their working lives.
Yet in 2025, 22 per cent of American 15-year-olds failed to reach this benchmark in science (up from 19 per cent in 2018), 35 per cent in mathematics (up from 27 per cent) and 26 per cent in reading (up from 19 per cent). These students struggle to interpret simple data, identify evidence in a graph or understand simple texts.
The picture is similar in many countries. Indeed, the world is no longer sharply divided between rich and well-educated nations and poor and badly educated ones. National wealth accounts for just half of the variation in countries’ performance, indicated Pisa.
In the same vein, good policies and wise spending choices are far stronger predictors for learning outcomes than just how much countries spend per student.
Luxembourg spends US$288,521 per student from the beginning of primary school up to the age of 15, by far the most, but it still posts lower science scores than Ireland, France or Italy – all of which invest less than half that amount per student.
Students in the Chinese jurisdictions of Beijing, Shanghai, Jiangsu and Zhejiang topped the Pisa science test, with all more than 100 points above the Organisation for Economic Co-operation and Development (OECD) average, where 20 points typically equal a school year’s progress.
Even if they represent just a small part of China, they make the largest contribution (27 per cent) to the global talent pool of Pisa’s top performers.
These students can tackle challenging scientific concepts, evaluate claims and identify flaws in scientific arguments, making them prime candidates for tomorrow’s top jobs. The education systems next on the list – Singapore, Macau, Taiwan and Japan – are all from East Asia, too.
The greatest cause for concern emerges not from a single snapshot of Pisa, but from the direction of travel.
At a moment when artificial intelligence is transforming how humans work, learn and live, one would expect education systems to be raising the bar. Instead, in many places, outcomes have declined.
Reading, the gateway to all learning, has declined sharply. Between 2018 and 2025, performance fell in three-quarters of the countries. Across the OECD, it was down by 25 points – more than a year of learning. Much of this decline happened between 2022 and 2025.
The biggest surprise is who fell furthest behind. It was not the most disadvantaged students, but those from wealthy families, which points to a much broader challenge than inequality alone.
“Few countries would allow millions of children to receive a new vaccine without first demanding rigorous clinical trials. Yet in education we introduce new technologies into millions of classrooms with remarkably little evidence about their impact.”
Most troubling is the deterioration of the very skills that matter most in the AI age: evaluating information, making connections across multiple sources and thinking critically about what we read.
Since 2018, the number of instances in which students rush through texts and give quick but incorrect answers has almost doubled.
People quickly turn to explanations such as the impact of the Covid-19 pandemic, but a decline in reading was already visible among 10-year-olds back in 2016, well before the virus hit.
We cannot say for sure, but falling levels of reading for enjoyment and the shift towards digital reading – skimming social-media feeds and rapidly processing information – may be reducing the ability of students to engage with complex texts and data.
Although limited or moderate use of digital devices for learning at school is often associated with better outcomes, when personal devices are used for leisure during school, student performance falls sharply.
This raises questions about whether devices such as smartphones, designed to maximise a user’s attention, are compatible with sustained academic attention.
Is the arrival of AI changing the equation? At first glance, Pisa data is not positive. Students who use chatbots for tasks such as summarising texts tend to get lower scores than those who do not, and are more likely to report lower levels of intrinsic motivation for learning.
The patterns are complex, and depend a lot on how AI is used, but the bottom line is simple – we do not get fit by watching sports, but by doing sports. In the same way, learning is always about cognitive struggle.
Rather than make things easier, we need rigorous, focused and coherent learning environments that make students stronger.
In Pisa, both individuals and countries perform better when students believe that success is about hard work rather than talent, and when they see failure as an opportunity for growth.
So how do we equip students for an AI-driven future when the same technologies can hollow out learning? The answer is not to reject AI, but to rethink how we use it.
High-performing systems such as China, Estonia, South Korea and Singapore are leading that shift, deploying AI not to give students ready-made answers, but to help them think about the right questions, reason analytically and learn more effectively.
Few countries would allow millions of children to receive a new vaccine without first demanding rigorous clinical trials. Yet in education we introduce new technologies into millions of classrooms with remarkably little evidence about their impact, turning students into unwitting crash dummies for tools we do not yet fully understand.
Pisa shows that schools with clear policies on mobile-phone use – whether through bans or well-defined guidelines – report better outcomes and fewer students distracted by digital devices.
The lesson is not that technology has no place in schools. It is that technology needs governance.
Where smartphones and AI are part of the learning environment, their use must be intentional, not incidental. Schools need clear frameworks defining when, where and how AI should be used.
For younger learners, it should not be treated like an oracle. That means strong filters. No emotional manipulation. No pretending to be a friend, a confidant or, worse, a replacement for social interaction.
Not least, without sustained investment in teacher skills, educators risk either outsourcing too much thinking to AI or rejecting innovations that could genuinely enhance learning. ©2026 The Economist Newspaper Limited. All rights reserved
The writer is director for education and skills at the OECD
