There is a comforting story going around about AI and education. It runs roughly like this: now that the machine can write the essay, do the maths, and structure the argument, the case for drilling children in literacy, numeracy and reasoning is fading. Teach them to use the tool, and the rest will look after itself.
It is a tidy story. The evidence from the first three years of working with these systems suggests it is also wrong, in a more interesting way than the headlines suggest.
In routine work, AI is a leveller
The most-cited piece of evidence on this question came from Erik Brynjolfsson and colleagues at Stanford and MIT, who studied more than five thousand customer-support agents using a generative AI assistant alongside their normal work. Across the group, productivity rose by about 14% on average. The gains were not evenly distributed. Newer, less-experienced agents saw their productivity jump by roughly a third. The most experienced agents saw essentially no change at all.
This is the part of the story that gets quoted most often, and it is genuinely good news. AI raises the floor. A weaker writer, a less-confident maths student, a junior employee, a worker in a second language: all of them, today, have access to explanation and feedback at a level that was unimaginable a decade ago.
If a child were going to spend their working life in routine, scripted, mechanical work, this would be where the story ended. The machine catches you up. The basics matter less. Relax.
In judgement work, AI is an amplifier
Today's children are not heading into routine work. They are heading into the parts of the economy that the machine cannot do alone, where someone has to decide what is worth doing, frame the question correctly, judge whether the answer is any good, and take responsibility for it. In that kind of work, the picture inverts.
A separate 2023 field experiment by Fabrizio Dell'Acqua, Ethan Mollick and colleagues at Harvard Business School worked with consultants at the Boston Consulting Group. Half were given access to GPT-4, half were not. On tasks the AI was suited to, the AI-using group produced work that was about 40% higher quality and 25% faster. But on tasks just outside the AI's capabilities, the ones that needed real judgement, the group using AI did worse than the group without it. They had trusted the machine where they should have pushed back. The skill that separated the two halves of the AI-using group was not access to the tool. It was the ability to tell when to use it and when not to.
AI multiplies what a child already is. It does not create what is not there.
That is the pattern teachers are now seeing in classrooms too. The pupil who reads carefully can interrogate a long answer and notice a contradiction; the pupil who skims will be skimmed back at. The pupil with real numerical sense can spot a wrong calculation in an instant; the pupil without it accepts the number on the screen. The pupil who has learned to push through a hard problem will use AI as a sparring partner; the pupil who has been protected from struggle will hand the difficulty over to the machine the moment it appears.
In every one of those pairings, the same tool produces wildly different work. AI does not flatten ability in skilled work. It multiplies it.
What gets multiplied
The fundamentals that AI most powerfully amplifies are not exotic. They are the ones a good school has always cared about.
Reading deeply. A child who can hold a long passage in mind, distinguish argument from assertion, and notice where the writer is bluffing can use AI to read further and faster. A child who skims will be told what to think.
Numerical reasoning. A child who has internalised quantitative thinking can interrogate a chart and smell out a wrong calculation. A child without it accepts what the screen says.
Writing as thinking. A child who has learned to think through writing, drafting, cutting, restructuring, can direct AI to produce work that genuinely reflects their ideas. A child for whom writing has always been mysterious will produce work that reflects the model's defaults, not their own mind.
Domain knowledge. A child with real knowledge of history, science or literature will ask better questions and recognise better answers. A child without it cannot tell brilliance from confident nonsense, and AI, fluent and persuasive, produces both.
The habit of hard thinking. Possibly the most important. A child who has been allowed to struggle through difficult problems will use AI as a partner in tackling harder ones. A child whose schooling has spared them the struggle will discover that AI is the perfect way to spare themselves it forever.
The strategic reading, for parents
Schools are sometimes asked, with a slightly nervous tone, whether the rise of AI means the basics matter less. The answer is the opposite. The basics matter more than at any point in living memory, because they are now multipliers in a way they have never been before.
A child with strong reading and strong reasoning, walking into a workplace where AI handles the routine cognitive work, will find that the parts of work only they can do, judgement, taste, the right question, the next step, are precisely the parts on which their value rests. A child without those things will find themselves competing with the AI for the routine work the AI will increasingly do better and faster.
The temptation to think AI levels the playing field is understandable, and it is generous. The first three years of evidence suggest the truth is more demanding. AI has raised the floor everywhere, and that is wonderful for a great many children. It has also raised the ceiling much faster. That ceiling is where the most interesting working lives of the next thirty years will be lived.
ISJ tracks pupil progress across the years through GL Assessments, with results published on the Results page. The work the school does on teaching quality connects directly to the argument here: in an age when AI multiplies what a child already is, the quality of the teaching that builds those fundamentals matters more than at any point before.