The AI Revolution — History Repeating Itself

A personal view of the British system — and why education keeps repeating the same response to every new tool

I am tackling education from a personal perspective; there is no AI Committee behind this one. I came up through the British world of O Levels and A Levels in the 1980s — secondary school, sixth form and then university. I was a STEM student, taking mathematics (pure and applied), physics, chemistry, biology and computer studies, and where the narrative allows I have drawn on my own experience.

I am also focusing on secondary education and above. Primary education should be about building the foundations on which all later learning rests, and I will leave that stage to one side.

An old worry with a new name

The current anxiety in education is not new. Each time a new cognitive tool appears — one that promises to do some of the mental work we previously did unaided — education reacts in much the same way. It first treats the tool as a threat, and then, often years later, adapts by redrawing the line between what must be learned without help and what may safely be delegated.

The move from "strictly forbidden" to "expected practice" is rarely a planned transition. More often it is a series of uncoordinated steps that leave teachers on the ground to work out — or quietly ignore — how to cope. Little thought tends to be given to those struggling with the change: rather than supporting them or bringing in people trained for it, the system imposes the shift and hopes for the best.

The pattern is far older than the computer. In Plato's Phaedrus, Socrates worries that writing itself will erode memory and hand students "the appearance of wisdom" in place of the real thing — pupils who can recite but cannot think (Plato, Phaedrus, via Wabash College). He had a point. Yet no one would give up writing to keep their memory sharp. That is the shape of the whole story: a tool arrives, something genuine is lost, a great deal more is gained, and education eventually rewrites its own rules around it. AI is only the latest name for a very old worry.

The assessment problem

Assessment is where the difficulty is most immediate. The assignment — long treated as a window into a student's thinking and reasoning — no longer reliably indicates whose thinking it reflects. With AI available to every student, submitted work can no longer be assumed to be the student's own.

The Higher Education Policy Institute's 2026 student survey, conducted with Kortext and polling just over a thousand UK undergraduates, found AI use to be "near universal": around 95% now use AI in at least one way, and roughly 94% use generative AI to help with assessed work — up from 88% a year earlier and just 53% the year before that (HEPI, 2026; HEPI, 2025). The same series shows the share of students pasting AI-generated text directly into assessed work climbing year on year — reported at 12% in 2026, up from 8% in 2025 and 3% in 2024.

These figures are not evidence of an unprecedented collapse so much as the opening of the latest chapter in a familiar story. We have been here before.

A short history of tools education feared

Memorising for its own sake

In my day, committing formulae to memory was often treated as a mark of academic discipline. I would ask my tutors under what real-world conditions memorising a formula would ever be the practical option. Some were candid enough to admit the exercise was pointless; others defended it, though the defence always came back to the same point: in the exam, you have to recall them from memory. Their justifications, like most of what I crammed, left my memory the moment the pass was in hand.

Calculators and clocks

The calculator, and later its programmable cousin, was another forbidden object; the two devices above are like the ones I owned. At first they could not be used in class: the fear was that students would lose their arithmetic fluency, grow dependent on the machine and stop understanding numbers. In time the rules softened — some papers permitted calculators while others did not — though my recollection is that programmable models stayed banned well into sixth form.

The more recent concern that some students cannot read an analogue clock rests on a real, documented trend rather than an internet myth. In 2018, UK schools reported replacing analogue clocks in examination halls with digital displays after teachers found that a noticeable number of GCSE and A-Level students struggled to read traditional clock faces quickly under exam conditions (The Telegraph; The Independent; BBC News). The likeliest explanation is a generation raised on digital displays having little daily need for analogue faces. The evidence does not support the stronger claim that a whole generation cannot tell the time; it points instead to a single skill weakened through lack of practice — much as earlier generations were said to lose mental arithmetic to the calculator, or attention to spelling once the spell checker arrived.

A weakened skill, though, is not a lost one. Any student who wants to understand how an analogue clock works now has endless free material to hand. We all leave school without truly grasping some concept or other; someone who never understood Newton's third law of motion can today find hundreds of explanations online, in a variety of styles — a world away from the single textbook, the scribbled blackboard and the occasional lab that were all I had.

Handwriting and the spell checker

Then there was writing itself. Joined-up, cursive handwriting was bound up with notions of neatness, maturity and educational seriousness. The spell checker drew two objections: that it would erode a person's ability to spell and communicate, and that by changing words it altered the writer's original text.

A trace of that caution survives in the exam hall. Under the rules of the Joint Council for Qualifications (JCQ) — the umbrella body that sets common exam regulations for the UK's main awarding organisations, including for GCSEs and A Levels — a candidate approved to use a word processor normally has spelling and grammar checkers and predictive text switched off, unless a further approved access arrangement allows otherwise. The machine is, in effect, turned back into a plain typewriter to level the field. Outside the exam room the norm has reversed entirely: in higher education a standard spell checker is simply expected, and coursework is meant to arrive clear, professional and free of errors.

Film, radio and television

If any medium was ever oversold to education, it was film. In 1913 Thomas Edison declared that books would "soon be obsolete in the public schools," predicting that the motion picture would teach every branch of knowledge and overhaul the school system within a decade (Quote Investigator). Radio, and later television, inherited both the hype and the suspicion.

The suspicion ran along familiar lines. Many teachers and traditionalists were sceptical, and their objections have a distinctly modern ring:

students would become passive observers rather than active learners;
entertainment would crowd out serious study;
teachers would lose their authority;
pupils would lean on images instead of developing imagination and reasoning; and
the technology would prove a distraction rather than an aid.

Neither the hype nor the fear was borne out. Film did not replace the textbook or the teacher, and television did not rot young minds. Once the novelty faded, moving pictures settled into the role they still hold — a supplement, valuable when a teacher uses it deliberately and hollow when it merely fills time.

The search engine

The web met the same reflex. When search engines put the world's information a keystroke away, the worry was that we would stop remembering anything and lose the capacity for deep, sustained thought — the argument in Nicholas Carr's 2008 essay "Is Google Making Us Stupid?" (The Atlantic). Two decades on, search is so woven into study and work that the question sounds quaint. The tool changed how we remember; it did not end thinking.

Wikipedia — the closest parallel

Of all these, Wikipedia is the closest parallel to how AI is likely to evolve in education, and its history is worth following in a little more detail.

When Wikipedia emerged in the early 2000s, much of the establishment reacted with suspicion. Its central idea — an encyclopaedia written and edited by essentially anyone — ran against academic notions of authority, authorship and peer review. Teachers warned of errors, vandalism and bias, and by the mid-2000s some schools and departments were explicitly barring students from citing Wikipedia in assessed work. Tellingly, these were usually bans on citing it, not on consulting it — Middlebury College's history department being a well-publicised example in 2007 (Inside Higher Ed). The case was complicated by evidence that it was not as unreliable as assumed: a much-discussed 2005 Nature investigation compared 42 science entries with Encyclopaedia Britannica and found Wikipedia surprisingly close in accuracy, though Britannica disputed the comparison (Nature).

Over the following decade the debate shifted from "is Wikipedia trustworthy?" to "how should it be used?" — as an entry point and a route to primary and secondary sources, rather than an authority to be cited in itself. British universities now reflect that more nuanced position, and some go further: the University of Edinburgh employs a Wikimedian in Residence and has students edit Wikipedia to build research and information-literacy skills, an approach it sums up as "don't cite Wikipedia, write Wikipedia" (University of Edinburgh). The response moved, in short, from prohibition, to reluctant acceptance as a starting point, to deliberate use for teaching source evaluation. Wikipedia never became a substitute for scholarly sources; the lesson was that students are better taught how, when and why to use it than simply told not to.

The pattern is consistent. Writing, the calculator, the spell checker, film and television, the search engine, Wikipedia — each was, in its moment, accused of hollowing out the mind. In every case the honest verdict is the same: something was genuinely lost, far more was gained, and education adapted once it stopped fighting the tool and started teaching it.

AI in the classroom

Much of the education establishment still regards AI warily — and not without reason: the models still invent things and state their errors with confidence. The picture is not uniform, though. Near-universal student use and real institutional change already sit alongside that unease: only around a third of students feel their institution encourages them to use AI, even as roughly two-thirds say assessment has already changed because of it (HEPI, 2026).

The reliability of the tools is improving, too. Grounding a model's answers in retrieved source material — an approach known as retrieval-augmented generation, or RAG — reduces some kinds of factual error, and having several models check one another can catch more. Neither technique is a guarantee: retrieval can surface the wrong passage, and models cross-checking each other can just as easily settle on a confident, shared mistake. These tools are becoming more reliable, not yet reliable — and teachers, too, are not infallible.

It helps to separate the ways AI can be used, because they are not equivalent:

As a tutor or explainer — standing in for or supplementing the teacher, the textbook, the library and the explainer sites, for instance by simplifying a difficult topic or summarising a long text.
As a generator of practice — producing quizzes to rehearse for an exam or a class discussion, or to confirm a topic has been understood.
As a maker of supporting material — infographics or images to accompany an argument, replacing the scrapbook cuttings and hand-built spreadsheet charts.
As a research tool — asked about Marie Curie, it can gather her life, family, work and achievements in one place.
As a reviewer — handed a finished assignment, it improves clarity, flow, spelling and grammar.
As a ghostwriter — the student pastes in the prompt, sets a length, an audience and a style, and submits the output, with varying degrees of review, as their own work.

As things stand, some of these are plainly unacceptable. Others might be permissible if educators could be confident that the models had guardrails to block what is not allowed and could not be jailbroken.

There are further reasons, not strictly about learning, why the sector is wary. AI is seen as environmentally costly — through the energy it consumes, the emissions behind that energy and its demand for water; it can be coaxed into explaining how to make poisons, weapons and other dangerous things; and it raises real data-privacy concerns, as students hand over personal information without realising it — pasting unreviewed text, feeding the model samples of their own writing so it can imitate them, or trading data for extra features. There is also an institutional risk: feeding assignments and graded work into a model can leak assessment material, making it harder to reuse questions and easier for the model to align its output with what markers expect. These are legitimate concerns the system must address.

What the computing era should have taught us

We have run a version of this experiment before. Computers arriving in schools produced the same spread of reactions. Some teachers were reluctant, having had little or no training; others embraced the machines and argued that programming and computer literacy were essential skills. Early adoption was constrained by shortages of hardware, training and suitable software (IJCSES). In Britain, the initial ambition was simply to put a microcomputer into every secondary school (Hansard, 1981) — which in practice often meant one machine shared among many pupils.

What eventually helped was a framework around the technology. National curricula defined what pupils should learn, and qualifications such as the European Computer Driving Licence (ECDL, later renamed ICDL) gave a standardised way to develop and demonstrate transferable skills for study and work (BCS). Meanwhile access broadened as computers, and then internet-connected phones, became ordinary: by 2018, nine in ten households in Great Britain had internet access (ONS).

Schools first concentrated their computers in dedicated suites, used both to teach computing and to support other subjects. As networks, laptops and classroom machines spread, those suites gave way to virtual learning environments and broader platforms that drew resources from many subjects together and let teachers, students and parents communicate, share material and track progress (Becta, 2010).

AI needs the same kind of plan. Blocking it wholesale is both wrong and a battle already lost — and, unlike the early computers, AI is already here and in every student's hand.

What we are actually assessing

The stated purpose of education in the UK has been remarkably stable: to develop the whole person and prepare young people for adult life.

In the 1980s, that meant proving what you knew without help. Information was scarce — the teacher, the textbook, the library, and if those failed you, cramming was the only recourse. I used to handle troublesome topics by dumping everything onto paper the moment an exam began, so I could refer back to it. The exam was the main event.

Today the task is almost the reverse: proving what you can discover, understand, evaluate, create and defend in a world where help is everywhere. The trouble is that the system is not built for it. The pressure to produce high-flyers is immense, and where income depends on reputation and recruitment — most obviously in the private sector and in universities competing for students — a slide in results can cost an institution applicants and money. Parents, reasonably, want their children employable or able to progress further in their studies, and in both cases the proof is a certificate — to show an employer or the next institution. Employers must sift hundreds of candidates in little time using those same certificates — and then complain that the graduates arriving are out of step with what industry needs.

If we could start again

Consider how the system might look if it were designed for this environment from the outset.

The pre-secondary years would concentrate on teaching children how to find and understand information using the available technologies in controlled settings, and how to organise their thinking with tools such as mind maps, flowcharts and concept maps, adapted to their age. Learning would lean on lived experience: in the absence of the environmental simulator from The Orville, schools could book television-style sets and spend time living a topic — running a small corner shop, say, to learn about trade, handling money, stocking shelves and selling. Ten- and eleven-year-olds already learn to do sophisticated things with their devices because they enjoy it; schooling should build on that same motivation. Children would be graded not on recall but on presenting from notes, debating and fielding challenging questions.

In secondary school, the flipped classroom would come into its own. Students would first meet the material on their own — controlled AI, video and reading of their choosing — having already been taught how to seek out, record and absorb information. They would arrive not to receive a lesson passively but to ask questions, practise, solve problems and get help. Traditional teaching would remain on hand for those who are struggling or who prefer it, and because pupils learn at different rates, peer discussion and staff support would smooth the sticking points. Wherever possible, learning would be hands-on. Each assignment would end in a submitted piece of work, checked against three questions:

Does it contain errors, whether human or AI?
Can the student demonstrate genuine understanding through a viva — a live spoken and written exchange?
If not, they get the chance to redo it using the resources available.

In their final two years, students could take up voluntary, vetted summer placements at approved employers for a taste of working life — with hours, pay and conditions set carefully to prevent exploitation. Employers would gain a first-hand view of whether the system is producing what they need, and a channel to say so.

A reform on this scale needs stakeholders who are committed rather than pompous, willing to hear other views and to set aside their own agendas. It should not be decided by whoever is closest to the minister's ear or has a product to sell, but by whose vision stands up to scrutiny. Mistakes will be made; once a direction is chosen it belongs to everyone, to be corrected and refined as it goes. Anyone who has used generative AI since its widespread public arrival in late 2022 knows how far steady refinement can carry a thing in a short time.

Two principles matter most. First, education should build on open-source software and open-weight models rather than depend on private companies whose duty is to their shareholders. The script is familiar: establish dominance, engineer lock-in, then monetise ever harder — degrading the free product until users pay for "premium" features, charging students for access, or mining their data for value, with a floor of lawyers to fight off anyone who objects. Open systems keep schools, and students of every means, in control of their own learning.

Second, fighting AI because the establishment cannot yet see how to work with it is exactly the mistake this history warns against. Educators with genuine, well-founded concerns about AI's harms are right to voice them — but the target should be the vast, under-taxed corporations powerful enough to face down governments, not the students who will have to live and work in the world these tools are building.

Education has met the writing tablet, the printed book, the calculator, the film reel, the television set, the encyclopaedia and the search engine, and feared each in turn. Each time, once it stopped fighting and started teaching, it emerged stronger. AI is unlikely to be the exception, provided the sector chooses to teach it rather than resist it.

Some will read this as a vision, others as wishful thinking or a fantasy better left to science fiction. Yet ideas dismissed as impractical have often, in time, proved to be exactly what was needed. As George Bernard Shaw put it, "the reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man" (Man and Superman, 1903). If a single idea here reaches someone with the standing to develop it, and the will to see it through for the benefit of students and society, then setting it down will have been worth the effort.

Follow This, That and (Maybe), the Other:

Search This Blog

TT(O)M