When I was growing up, my dad and I often talked about a story we wanted to write together. It was called “The Day Nobody Went to Disneyland”. One day, the story goes, the weather is so perfect that everyone in the world decides to stay away from Disneyland because they expect it to be far too busy. But my dad, who has the same thought as them, sees the disappointment on my face that morning – and decides to risk it. We set off. And when we arrive, the park is completely empty. Nobody else was brave enough to visit. We have everything – the rides, shows, food – to ourselves.
I loved the idea of writing that story. But, as so often happens, we were too busy to do anything about it. Life ran away from us. I grew up, and we both forgot about it.
Thirty years later, I was at my parents’ for dinner with my own children (eight, five and two). For some reason, the idea popped into my mind again. I told my oldest about it – and she wanted to hear the full thing. We turned to ChatGPT, gave it the vague outline and asked it to have a go at writing the story in the style of Dr Seuss (one of our favourites). Within a few seconds, we had it. She read it out – and we laughed.
Of course, what this AI generated wasn’t perfect. A few lines didn’t quite make sense; a few Americanisms had slipped in (“sneakers”, “cinnamon buns”). But that night on the sofa we had great fun together tweaking the prompts, refining the story, adding more scenes, working with this strange new creative partner to do something that had escaped my grasp for decades.
For the last 15 years, I have been exploring the impact of AI on work and society. But becoming a dad of three small people has made these disruptions far more personal: it is clear that their future is going to look wildly different from my past.
For my wife, that realisation really sunk in a few weeks ago, on a journey to her parents’ in Suffolk. Three children, confined in a car, stuttering along the A12, is a combustible setup. So we were thrilled to find a rare podcast that everyone miraculously enjoyed: History’s Not Boring, a series of 15-minute bursts of conversation, on a kaleidoscope of subjects, narrated by two children.
As we listened, we started to speculate in the car about who these whiz-kids might be. Where were they from? How were they selected? How did they juggle this job with school? Eventually, I went to the website to find out. All our answers were wrong. The kids did not, in fact, exist. The entire podcast seemed to be generated by AI. None of us had realised.
It was a strange moment, learning that people we had grown fond of were not actually people at all. It was also remarkable that AI could create such high-quality educational content. But for my wife, a podcast and documentary maker, it was an unsettling moment as well: here was something that would have taken her and a talented team several days to make, and now it was being done without people.
My work on AI has taken me to boardrooms and conference halls, classrooms and government buildings. But wherever I go, one question is now asked more than any other: what should my children do?
Today, we are floundering. Teachers worry that the way they teach no longer seems to work. Parents, watching their children use AI to solve problems and answer questions, wonder what they really know. Employers doubt whether traditional educational achievements, awarded in time-honoured ways – coursework, exams, assessments – are still a useful guide to what young people can do.
Yet reacting to these new technologies by putting barriers up, as many instinctively want to do – banning, admonishing, punishing – just cannot be right. To begin with, this is the water in which the next generation will swim: we are letting them down if we prepare them for the world that we grew up in, not the one they will actually inhabit.
How often, for instance, have you heard people accuse students of cheating if they use AI to help them with their work? But at what point are we cheating them by failing to overhaul the way we are educating them? If AI is making education “too easy” for young people, at what point does the burden shift on to us, the adults, to make their education more testing, to push them on, to stretch them?
More importantly, this dismissiveness of AI is also unimaginative. It stops us appreciating the astounding possibilities AI could create for the next generation as well. Of course there are risks. But at the same time, if we use this technology wisely, what difficult ideas might the next generation now grasp, ones that were beyond us when we were in their shoes? What hard problems might they now solve, ones that once required years of training and experience – if they were solvable at all?
Forget future-proofing skills
A big part of the challenge of AI is that our traditional response to technological disruption in the working world – the idea of “future-proofing” people – no longer works. In 2013, the then prime minister David Cameron announced that England would become the first place in the world where all children in primary and secondary schools would learn to code. This, in the words of education secretary Michael Gove, would “equip every child with the computing skills they need to succeed in the 21st century”.
The idea seemed bold and clever: in the years that followed, it was hard to find an advanced country that did not follow our example.
Fast-forward to today. What does it turn out that the latest AI systems, such as ChatGPT and Claude, are best at doing? Writing code. In January 2026, Anthropic reported that 90% of the code for Claude Code – their AI-powered coding assistant – was itself written by AI. Skills that were meant to protect kids from technological disruption for their entire lives were largely redundant before they had even left school.
For politicians and policymakers, it is tempting to dismiss the coding saga as an unfortunate blip. But this is a mistake. The reason they slipped up was because they had believed in the idea of “future-proof” skills, the thought that there exists some valuable set of skills that AI will not be able to do for some time – and that by thinking deeply enough about the future, we can accurately identify them.
The truth is that we know only two things about what lies ahead. One is that it will be full of technologies that are far more capable than today. And the other is that we know little else. But rather than attempting to resolve this uncertainty, we have to accept it and instead ask a different question: how do we prepare the next generation to flourish in a future that we actually know surprisingly little about?
Get back to basics
The first step is to get back to basics. Since 2009, literacy and numeracy have been falling among young people around the world, according to the OECD’s programme for international student assessment (Pisa); the same holds true for adults. In itself, this is worrying. But given the uncertainty that we face about the future, these trends are a disaster.
Why such a calamity? To begin with, while it might be hard to say precisely which more advanced skills will turn out to be most valuable in the future – creativity, judgment, something else – they will rely in some way on these basic skills. They are the foundations for everything else.
What’s more, AI systems are far from flawless. They make mistakes, often on simple problems; they hallucinate, providing confident and plausible answers that are completely made up. They are, as the computer scientist Geoffrey Hinton put it, “idiot savants”. That means we must use AI critically, not blindly, keeping our basics sharp so we can tell when AI is being a savant – or an idiot.
So, with both those reasons in mind, we should be focusing intensely on teaching literacy and numeracy, even if AI does them better, as it increasingly seems to do. Getting back to basics is what an economist would call a “no-regrets” strategy – we will never regret improving these basic skills, however the future turns out to be.
And we should try to be imaginative. When my eight-year-old grew bored learning her times tables the traditional way – long lists, learned by rote – I turned to ChatGPT to create various computer games, designed by her, to test what she knew. (Unicorns and rainbows featured in all versions.)
Then there was the time my five-year-old was tired of reading his standard-issue phonics books after school – there really is only so much Biff, Chip and Kipper a human being can take – and instead we used AI to generate some custom stories together, carefully tuned to his reading level and much closer to his favourite topics. (At the time of writing, HMS Belfast, TNT explosives and Bukayo Saka.)
Experimentation is important and we can learn so much from the creativity of others. Take Chris Moran, for instance, head of editorial innovation at the Guardian. His daughter had been reading Dracula at school but was struggling to place the sprawling novel in the real world. Together, with the help of AI, they built an app – PlotLines – which placed the story on an interactive 1890s Ordnance Survey Map, plotting the key scenes and character journeys around Europe, with the help of AI. (They have done it for many other books now, too.)
These are precisely the sorts of innovations we should be testing and embracing throughout education – rather than banning.
Teach both, test both
Despite the uncertainty we face, there is still one thing that we do know about the future: it will be full of technologies that are vastly more powerful than today. With that in mind, we must teach the next generation to use them. To do this properly will require us to dedicate a serious chunk of time to teaching people how to use AI.
The challenge, though, is how to do this without also setting them up to forget crucial basic skills. An important, growing fear is that AI is making us stupid. Why bother reading a book if AI can summarise it? Why bother completing maths homework if AI will do the sums? And in my view the answer here lies not in Silicon Valley, but with a forgotten British maths professor, Wilfred Halliday Cockcroft, half a century ago.
In the 1970s, the quality of maths teaching in UK schools appeared to be collapsing and numeracy was reportedly low. Cockcroft, who had a longstanding interest in mathematics education, was asked by the government to look into it. And in 1982 the Cockcroft Report was published. It was enormous, forensic – and is now largely forgotten. But it may turn out to be the most important document written for thinking about the future of education, because of its response to the electronic calculator.
It is fascinating to read Cockcroft’s report and discover how the challenges of the calculator back then were so similar to the ones we face with AI today – from the fear that it would undermine “basic skills” to the intimidation many felt at the complexity of this strange new technology (“Some … had been discouraged by the large number of figures which had appeared after the decimal point”).
Cockcroft was realistic: “all candidates”, he expected, “will have access to a calculator by 1985”. And his proposal was revolutionary: overhauling mathematics education by splitting it into two parts, spending part of the time teaching students to use a calculator and the rest of the time learning to cope without one – and crucially, testing both. It caught on. Learning maths without and with a calculator is now the gold standard around the world, the former nurturing the basics and the latter applying them to more fiendish problems.
We should also be realistic and revolutionary, adopting this tried-and-tested principle – what I call “teach both, test both” – for AI now. Every subject, from history to English literature, should be divided in two: teach students to use AI in one part, teach students to flourish without it in the other – and, crucially, examine both. A teacher cannot monitor whether or not a student uses AI in the quiet solitude of their bedroom. But nothing can replace the feeling of sitting in an exam, looking at the paper, and feeling that cold sweat when you realise you haven’t prepared for both parts.
Not all screens are bad
One of my worries is that we allow a legitimate sense of concern about the impact of social media on children’s lives to seep into how we think about AI, allowing good restrictions on the former to turn into kneejerk bans on the latter. Social media and AI are not the same thing, and whereas the former often dehumanises and distracts us from the real world, AI – used in the right way – can make our lives go far better.
There is another way to think about this: a debate is now under way about the “Goldilocks” amount of screen time for children – not too much, not too little, just the right amount. But this is not the right argument to be having. What matters is not so much the amount of screen time, but what is actually on those screens, what we are using this technology to do.
This distinction matters not only for thinking about what we teach, but how we teach as well. Consider a further example: personal tuition. It is often said an average student who receives one-to-one tuition will outperform almost all their peers in a traditional classroom setting. I saw this first-hand, as a tutor in Oxford for years, teaching mathematics and economics.
The problem, though, is that human tutors are too expensive to provide to everyone. But now, AI can finally provide high-quality personal tuition, tailoring the way material is taught to the unique strengths and weaknesses of each student, mimicking interactions with a human tutor but at a far lower cost.
In all honesty, AI provides a level of tailored instruction that I – and many other teachers I have watched over the years – have struggled to achieve. In part, it is the breadth of AI that is striking. I have used it to respond to my five-year-old’s bedtime-delaying tactic of asking vast questions just as I turn down the lights (in response to, “Daddy, where did the first human come from?” it drew up a short story about evolution), as well as to create step-by-step instructions to help graduate students solve hard mathematical problems in economics, such as the Ramsey growth model.
“I don’t think anyone has ever paid such pure attention to me and my thinking and my questions,” a student was reported as saying in The New Yorkerin April 2025. “It’s made me rethink my interactions with people.” AI never gets tired or distracted. It doesn’t have fixed office hours or limited classroom time. It will always answer one more question, always provide one more explanation of a problem you don’t understand.
Focus on problems, not the job
I am not surprised that students have been booing tech titans during recent US graduation ceremonies. The most resistant groups I have spoken to over the last 15 years are young professionals: they have spent a big chunk of their life, and sunk huge amounts of money, preparing to be a lawyer or doctor or whatever it might be, and they are understandably furious when they are told, just as they cross the finish line, that the world they were preparing to enter is over.
There is so much advice I would want to give them, worried at the start of their careers. But one of the most important bits is to choose a profession because the problem interests you, not the job. Bluntly, if you go into medicine because you like the look of doctors on House, or law because of Suits, or marketing because of Mad Men, then you are going to be disappointed. These jobs are soon going to look very different.
But the problems themselves – improving health outcomes, providing legal advice, selling products? They are not going away. It is the way we solve them, and the skills required to do so, that will look very different.
You could glimpse this well before the arrival of generative AI. Back in 2017, a team at Stanford University announced they had built a system that could tell whether or not a freckle was cancerous from a photograph as accurately as leading dermatologists. It was a remarkable moment, a lurch forward in AI-enabled diagnostics. Yet what was particularly interesting about this work is that the final co-author on the Nature article that announced this achievement was Sebastian Thrun – not a medical doctor, but a leading computer scientist who developed the world’s first driverless car. Here was a man who knew very little about medicine at all, yet with the skills that he had was able to build a system that could rival the expertise of the finest doctors.
Run towards AI and science
What would I actually do if I were at the start of my career, back in the starting blocks of life? Without hesitation, I would run towards AI and science – not because it is protected from automation, but because it is where the most excitement is likely to happen in years to come. In the 20th century, the best ideas we had about the world came from the heads of clever human beings. In the 21st century, I expect they are likely to come from capable AIs instead.
We could catch a glimpse of that in late 2024, when the creators of an AI built by DeepMind – AlphaFold2 – won the Nobel prize in chemistry for solving the “protein folding problem” (one of the greatest unsolved challenges in biology, critical for understanding how diseases work and how to treat them). At the time of writing, frontier mathematics is next in line, new discoveries appearing at an impressive rate.
Imagination, imagination, imagination
The more time I spend with AI, the more I realise the main constraint on its use in the world is the limits of our imagination. Even if we were to press pause on progress in the technology today, there would be vastly more uses for it than we have currently dreamed up.
Part of the imaginative task falls to us. But that night with my eight-year-old, crafting that story, reminds me that part of it falls to the next generation as well. “We look at the world once in childhood,” wrote the poet Louise Glück, “the rest is memory.” In that spirit, we need the next generation, with their sense of adventure and open-mindedness, their lack of world-weariness, to help us think wisely and freely, together, about what we can use these extraordinary technologies to do.
