Who’s Supposed to catch a Generation that Outsources its Thinking?

I read three things recently that felt worth pulling together, if only to highlight them rather than fully make sense of them.

The first was two weeks ago PISA headline: stability. England’s 15-year-olds held their ground in maths, nudged up in reading, improved in science, while most of the OECD went backwards in the first two. We should take reassurance in this, and I’m sure it will give pause for thought to a new ministerial team.

As in every PISA cycle, in every OECD country, pupils from disadvantaged backgrounds scored well behind their better-off classmates in all three subjects. The science gap across the OECD runs to 85 points. Where it narrowed slightly this round, the OECD’s own analysis says that’s because advantaged pupils slipped back, not because disadvantaged pupils caught up. These are also the pupils least likely to have the resources at home to build knowledge outside the classroom, and most likely to need a system that catches them later if school doesn’t. Stable is a statement about the middle of the distribution. It says little about the cohort of young people we are most concerned about in 2026.

The Flynn Effect went into reverse

I also read a chapter published by Barbara Oakley and colleagues that documents something specific: IQ scores rose steadily through most of the twentieth century, the Flynn Effect, then began reversing in several wealthy countries, Britain among them, for people born from the mid-1970s onward. The timing tracks each new wave of what the authors call cognitive offloading, decade by decade: pocket calculators in the 1970s, then computers and internet search through the 1980s and 90s. The researchers are careful about correlation and cause. But the mechanism they describe is specific. When knowledge is only ever looked up rather than held and retrieved, the brain never builds the internal shortcuts that let someone recognise instantly when an answer is wrong. Retrieval practice builds fluency. Constant lookup doesn’t.

Which makes the PISA detail underneath the headline more interesting. England’s maths score in 2025 was still twelve points below 2018. And stable describes a cohort who grew up with smartphones and search from early childhood, then likely met generative AI, a far more powerful offloading tool than a calculator, only in the last two years of school. If Oakley’s mechanism stands firm, the group whose results would show it hasn’t sat a PISA test yet.

A smaller system to catch people later

Whilst I would be overzealous to conclude what this means, I’m willing to flag my current position as ‘alarmed’ if we move away from school assessment that encourages retrieval practice. Not least because, as reported by the Learning & Work Institute a fortnight ago, the system built to catch that generation if it stumbles is smaller than it was. Funding for adult English and maths, the route back in for anyone who leaves school without secure literacy or numeracy, has fallen by around 40 per cent over the past several years. Participation is down by half over the past decade. Less capacity to repair a gap after the fact, at the exact point a more powerful offloading tool has entered everyday teenage life.

I would argue that responsibility to get involved in the solution now sits a little more directly with employers than it did a few years ago. For example, when the government scrapped the level 2 English and maths exit requirement for adult apprentices in February 2025, DfE’s own data shows apprentices studying towards level 2 English and maths each fell by over 50%, around 65,000 fewer people working toward those qualifications than before the change. The old rule had real problems. It was failing capable apprentices on tests unrelated to their trade, and removing it wasn’t unreasonable. But it also removed the one structural nudge that made building those basics part of the job by default. Employer and apprentice can now simply agree it isn’t needed. That’s not a criticism of anyone making that call. It shows what the choice rests on: employers who keep English and maths in the training plan, even where it’s no longer compulsory, are doing some of the work this piece has been describing. Building fluency AI can’t substitute for, rather than assuming someone else already has.

The same shift in five different rooms

Of course, correlation across decades isn’t causation. One stable PISA round isn’t a verdict. But a documented historical link between offloading and cognitive decline, a national plateau hiding real strain among disadvantaged pupils, and a shrinking safety net at the exact moment employers have more discretion over whether to build these basics in at all is not filling me with confidence. It feels, more than at any point in my career, like something no single phase of education can answer alone. I’ve spent enough time moving between early years, schools, colleges, universities and workplace training to notice how rarely those worlds compare notes on the same problem. Each is having its own separate conversation about AI: what it means for exam integrity, for degree assessment, for on-the-job learning, for a child handed a tablet before they can read, for the young person entering the workplace who used ChatGPT a little more than we’d like for their coursework. It’s the same shift showing up in five different rooms.

By Lou Doyle, Director and Co-Founder of SDN Mesma Group