Near the end of an episode of Moonshots with Peter Diamandis, published on September 19, 2026, Peter Diamandis showed his co-hosts a chart of international test scores. Diamandis founded XPRIZE and Singularity University. The chart showed reading and mathematics scores for 15-year-olds holding roughly steady for years and then falling sharply. He called them "the steepest declines in the history of the test" and said "the problem is not in the future. The problem is right now."
The hosts then split over a basic question: if machines keep getting smarter, does it still matter when humans get worse at reading and maths?
What the chart measures
The scores come from PISA, the Programme for International Student Assessment, which the OECD runs. The OECD is an organisation of mostly wealthy countries. PISA tests 15-year-olds in reading, mathematics and science. Diamandis said the fall began before ChatGPT arrived and was "probably due to smartphones and social media that did the first damage." He also read out a question from Vala Afzhar, who had posted the chart: what happens when we stop reading and writing and start outsourcing our thinking?
The OECD's own reports give a more mixed picture than a single falling line. Its analysis of the 2022 results found that between 2018 and 2022, average mathematics scores across OECD countries fell about 15 points and reading about 10 points. Science did not change significantly. Those drops were unusually large, and they happened before ChatGPT was in wide use. The OECD links them to pandemic disruption and to longer-running problems in education. Earlier trends were not entirely flat either. On average, reading had peaked in 2012 and science in 2009. Some systems moved the other way: Colombia, Macao, Peru and Qatar improved in all three subjects over the years they took part.
The PISA 2025 results, published on September 8, 2026, cover more than 760,000 students in 91 countries and economies, representing about 33 million 15-year-olds. From 2015 to 2025, the OECD average fell 22 points in mathematics and 28 points in reading. Science fell more modestly.
This round also asked students about AI, and the answers depended on how they used it. Students who used AI for specific schoolwork tasks scored about 20 points lower in science than students who did not use it. Students who used general-purpose AI tools every week scored about the same as non-users. General users who had opportunities at school to learn AI literacy, meaning how the tools work and how to use them well, scored slightly higher, but advantaged students were more likely to get those opportunities. These are associations. They do not show that AI caused any of the differences.
Comparing countries or years also depends on who gets tested. According to the OECD's sampling rules, PISA first selects schools and then randomly selects students within them. It tests enrolled students aged about 15 in grade 7 or above, so students are not all in the same school year. Excluded students, students who do not take part and differences in who is enrolled in school can all affect comparisons, and not every system met all of PISA's participation standards.
"So who cares?"
Salim Ismail spoke first. He founded Open ExO and is a partner at Exponential Venture Capital. He said losing critical-thinking skills is "a real problem" but offered another reading. For centuries, he said, people trained their minds for deep technical work such as mathematics, accounting and engineering. As AI takes over more of that work, the valuable skill may become the generalist's: deciding which problems are worth thinking about. He noted that more than half of Silicon Valley CEOs have liberal arts degrees. He called the decline "a natural progression" and said the response should be "much more verbal challenging around a student's understanding of topics", questioning students out loud about what they understand. Diamandis said he had been quoting Ismail's framing to friends: "the transition from human doing to human being."
Computer scientist Alexander Wissner-Gross, founder of Reified, went further. "The capabilities of frontier models are increasing far more quickly than the purported capabilities of humans are decreasing," he said. He doubted there was a problem at all. People in the industrialized West, he noted, don't worry much that they have lost the skills of farm workers. He predicted that humans will "merge with superintelligence in a few years" through wearable devices, brain-computer interfaces and full human-machine mergers. That, he said, makes falling scores at most a "two- to three-year problem" and a "mini-moral panic" about "outdated tests." He added: "So who cares?"
When Diamandis asked whether people will want to merge with machines, Ismail answered "of course." He argued that people wearing glasses or getting vaccinated are already using technology to improve on the human body, and that AI speeds up something humans have always done.
"This is bad"
Dave Blundin, founder of Link Ventures, said he saw it "a totally different way" and was surprised by his co-hosts. When reading scores plummet, he said, he does not think "who cares? It may not even matter. We'll all upload." His reaction: "No, this is like fundamentally horrible." He blamed social media, not AI.
Ismail pushed back. Literacy, he said, peaked around 200 years ago and has been sliding ever since, and more capable AI that summarizes and makes sense of things will speed that along. Fighting AI on raw thinking power is "a losing battle," in his view; what becomes uniquely human is the experience of being alive.
Blundin answered with a story from his own family. The headmaster of the school his children attended for middle school used to take the whole seventh grade to an island off the coast of Maine for three days with no phones. The students were aghast beforehand, Blundin said, but came back "the happiest they'd ever been in their lives" every year. The trip was later ended, he said, because it was considered inhumane. His worry is a generation that thinks constant texting is normal because it has never known anything else: "until you have, you don't really have an opinion."
Ismail had a similar story. His son goes to a summer camp that bans phones and comes home "the most joyous human being," he said, "for about three days until he's back in his devices." He passed on a line from a parent named Lily at a parents' event: when her child has a device she gets "20% of the kid," and without it, 80%. Ismail still concluded that technology will move forward and that people must "integrate with it, and then we find our humanity in that context."
Beyond test scores
Blundin closed the discussion by pointing to a risk beyond test scores. In the 1960s, he said, people discovered they could make drugs that everyone thought were fine until users came out "totally mangled for life," and electronics can do the same. He expects something to appear soon in the age of AI "as addictive as the worst drug ever created." The goal, he said, is "really happy, really healthy, infinite longevity people," not "the most addictive video game in the history of the world" that leaves people depressed.
The OECD's 2025 report also looks past test scores. It reports that students' sense of belonging at school improved, and it recommends that schools support motivation, students' sense of control over their own learning, families and teacher relationships alongside academic skills. On the podcast, the question stayed open: whether human skill and happiness are goals in their own right, or problems that smarter machines will soon make irrelevant.