There are statistics that sound bad and statistics that sound like the opening scene of a dystopian film. The latest results from the OECD's Programme for International Student Assessment belong rather uncomfortably in the second category. Across the developed world, 15-year-olds are performing in reading at approximately the level that would have been expected of 14-year-olds only seven years earlier. Mathematics has gone backwards as well. The great educational machine has apparently discovered reverse gear.
The immediate temptation is to blame COVID-19. Schools closed, children were sent home, lessons moved onto screens and the largest educational experiment in modern history was conducted without a control group. When the 2022 PISA results appeared, the enormous deterioration could therefore be explained, at least partly, as pandemic damage. Governments could tell themselves that the children had suffered an extraordinary interruption and that normal schooling would gradually repair the loss. That comforting explanation has just acquired a large hole.
PISA 2025 involved more than 760,000 students across 91 countries and economies. Across OECD countries, reading performance fell by 28 points between 2015 and 2025, which the OECD estimates as roughly one and a half years of learning. Mathematics fell by 22 points, equivalent to just over a year. One in five 15-year-olds is now classified as a low performer simultaneously in science, mathematics and reading, compared with 16 per cent in 2022.
More disturbing still is what happened after the pandemic. Average reading performance across the 35 OECD countries with comparable data fell 11 points between 2018 and 2022, but then fell another 14 points between 2022 and 2025. In other words, the educational patient did not simply take a terrible COVID hit and begin recovering. On this measure, the deterioration accelerated after the emergency was over.
That raises a much more disturbing possibility: perhaps we are looking at an educational ratchet. A ratchet moves easily in one direction but resists movement back in the other. Applied to education, the hypothesis would be that major social disruptions can permanently knock achievement downward. Schools subsequently adapt to the lower level rather than restoring the old one. Expectations change, curricula adjust, teachers accommodate weaker students, assessment standards move, parents become accustomed to poorer performance and the exceptional gradually becomes normal. The next generation therefore begins not from the old baseline but from the damaged new one.
Suppose a population of students loses approximately a year's educational development during a major disruption. That does not mean the next pandemic will automatically subtract another year. Human beings and educational institutions are not mechanical devices. But if the causes that prevent recovery remain in place, another sufficiently large disruption could push achievement downward again. Instead of a temporary V-shaped decline followed by recovery, society gets a staircase descending into the basement.
COVID may therefore matter less as the sole explanation for educational decline than as an experiment revealing how fragile the system had already become. The OECD's longer-term figures show that important deterioration predates the pandemic. Reading and science performance had already been trending downward in many countries. COVID landed on a structure whose foundations were already cracking. Now another suspect has entered the room: the screen.
The OECD's 2025 findings point towards declining reading habits, digital distraction and complicated effects from artificial intelligence. The problem is not simply that children possess computers. Computers can obviously be extraordinary educational instruments. The question is what happens when sustained attention is replaced by an environment built around notifications, short videos, scrolling, skimming and instantaneous answers.
Reading a difficult book requires remaining mentally attached to an argument for minutes or hours. The digital attention economy rewards almost the opposite cognitive behaviour: change the stimulus, click the link, watch the clip, check the message, ask the chatbot. A student can consequently become extremely proficient at obtaining information while becoming progressively worse at thinking through it.
Artificial intelligence adds another twist. Used intelligently, AI can function as tutor, critic, research assistant and intellectual sparring partner. Used badly, it can remove precisely the mental labour through which learning occurs. If the machine summarises the chapter, constructs the argument, solves the problem and writes the response, the student may obtain the product of cognition without undergoing the cognitive process that produces the ability. It is rather like sending a robot to the gym and expecting your own muscles to grow.
The OECD is careful about causation here, as it should be. Its data do not establish that AI or smartphones alone caused the decline, and some high-performing Asian education systems make extensive use of technology. What appears to matter is how technology is incorporated into education. Technology disciplined by education may enhance learning; education subordinated to technology may do something very different.
The deeper problem is that several downward pressures can reinforce one another. Reduced reading produces weaker readers. Weaker readers find demanding texts more difficult and therefore avoid them. Schools respond by simplifying material. Students consequently receive less practice with difficult material and become still weaker readers. Digital entertainment makes sustained reading less attractive, while AI makes avoiding difficult intellectual work increasingly easy. Each adaptation is individually understandable, yet collectively they can produce a feedback loop. This is where the ratchet becomes genuinely worrying.
Imagine another pandemic in ten or fifteen years. Perhaps schools close again for six months. By then a generation accustomed to AI assistance and highly fragmented digital media encounters another prolonged educational interruption. Governments again promise enormous catch-up programmes. But if the experience of COVID is indicative, the lost ground may not actually be recovered.
Then imagine another disruption twenty years later. Continue the process long enough and we eventually arrive at the satirical destination: Planet of the Apes. Not because chimpanzees suddenly acquire PhDs, but because the humans have misplaced theirs.
The joke contains a serious point. Civilisation depends upon accumulated cognitive capital. Every generation inherits mathematics, science, history, literature, engineering and institutional knowledge that it did not create. Maintaining a technologically sophisticated civilisation therefore requires each generation to acquire enough of that accumulated knowledge to operate, criticise and extend it.
We usually assume this transmission happens automatically, but it does not. A civilisation can accumulate knowledge while individuals become less capable of understanding it. Indeed, modern technology makes that paradox increasingly possible. We can build machines that remember more so that humans remember less, machines that calculate more so humans calculate less, navigation systems that know every road so that humans learn none, and AI systems capable of processing enormous bodies of literature while their users increasingly struggle to read a long article.
Society could therefore become simultaneously more technologically intelligent and individually less intellectually capable. That possibility is considerably more interesting than the familiar argument about whether PISA rankings prove that Country A has better schools than Country B. The latest results may be detecting something larger: a change in the cognitive ecology within which education takes place.
There are reasons not to surrender to educational doom. Decline is not universal. Some countries have resisted it, and the differences themselves demonstrate that deterioration is not biologically inevitable. If children in one educational culture can still achieve exceptionally high levels while comparable children elsewhere decline, something about institutions, expectations, habits or teaching practices must matter. That is good news because institutions and practices can be changed.
But first governments have to abandon the comforting assumption that post-COVID educational losses will naturally heal with time. They plainly have not healed yet.
The educational ratchet hypothesis therefore deserves to be taken seriously. COVID may have shown that modern educational systems can lose enormous amounts of human capital very quickly and regain it painfully slowly, if at all. Smartphones, declining reading, absenteeism, weakened academic expectations and poorly used AI may then prevent the pawl from releasing and allowing the mechanism to move upwards again.
If another great disruption arrives before those weaknesses are repaired, we may discover that the next step downward is easier than the climb back. No, Planet of the Apes is not around the corner. Caesar can leave his horse in the stable for the moment. But the underlying principle is decidedly less funny. Civilisations are not guaranteed to become progressively cleverer merely because their technology becomes progressively cleverer.
It is entirely possible to build smarter machines for a population becoming less capable of understanding them. And if the PISA trend continues for another generation, the most important educational question may cease to be why the children fell behind during COVID. It will be why, once they fell, we assumed they would automatically get back up.
https://www.oecd.org/en/publications/2026/09/pisa-2025-results-volume-i_5265bfb1.html
https://www.abc.net.au/news/2026-09-08/australia-records-worst-ever-reading-and-maths-results-pisa/107101504