Slot Machines in the Classroom: Who Must Prove Learning Apps Teach?

On a bright afternoon at a primary school in Hertfordshire, a headteacher did something quietly radical. She cancelled the afternoon's lessons, opened the back doors, and told her pupils to go outside and play. What happened next has since been recounted to audiences of thousands as a small parable of a large problem. The children did not scatter into games. They stood in the yard, uncertain what to do with an unstructured hour and an open sky. They could not readily invent their own entertainment. Within minutes, some of them asked to come back inside.

The actress and campaigner Sophie Winkleman told this story at ARC 2026, the London conference convened by the Alliance for Responsible Citizenship, in a talk provocatively titled “Why AI in the Classroom is a Catastrophe.” The talk circulated widely through the summer. Winkleman used the empty playground as evidence for a claim easy to state and hard to prove: that a childhood saturated in constant, low-level digital stimulation had eroded the very faculty a playground is meant to exercise, the ability to generate one's own amusement out of nothing but time, space and other children.

You can take the anecdote as a moral panic dressed in a good story, or as a genuine warning. Either way, it points at something specific and testable, and it is worth being precise about what that is. The worry is not simply that children spend too long looking at screens, a complaint as old as television. The worry is that a particular category of software design, refined for a decade inside the attention economy and increasingly restricted for use on children in that setting, has migrated into the tools schools use to teach reading, mathematics and languages. The mechanics that regulators have started to treat as harmful when a social platform deploys them to hold a child's gaze are frequently the same mechanics an educational technology company deploys to keep a child inside a maths app. And nobody, at the point these products are bought and installed at scale, is being asked to prove that the second use is meaningfully different from the first.

That is the question worth sitting with. If the engagement machinery that is driving measurable harm in children's social media use is structurally identical to the machinery now embedded in the software schools use to educate, what obligation do schools, regulators and the companies themselves have to distinguish a product designed to teach from a product designed merely to hold attention? And crucially, before these tools reach millions of children, who should carry the burden of proving the difference?

The mechanics that migrated from the casino to the phone

To understand why an educational app and a slot machine can share a family resemblance, you have to understand the small number of design primitives that make modern consumer software compulsive. There are essentially three, and none of them is new. What is new is their precision.

The first is the variable, or unpredictable, reward. The principle comes from operant conditioning research conducted in the mid-twentieth century, which established that a behaviour is reinforced most powerfully not when it is rewarded every time but when it is rewarded unpredictably. A pigeon that receives a food pellet on a random schedule pecks more persistently than one rewarded on a fixed schedule. A slot machine works on exactly this logic, and so does the pull-to-refresh gesture on a social feed, which delivers something interesting often enough, and unpredictably enough, to keep the thumb moving. The design writer and former Google ethicist Tristan Harris popularised the comparison years ago when he described the smartphone as a slot machine in every pocket. Even the market-leading language app Duolingo has been analysed, by designers largely sympathetic to it, as deploying uncertainty deliberately, so that on some questions a user loses a heart and on others does not, an ambiguity described in these analyses as mildly activating in the same way a slot machine is.

The second primitive is the streak, a running count of consecutive days of use that converts a habit into an obligation. The streak weaponises two well-documented quirks of human psychology, loss aversion and the sunk-cost fallacy. Once a user has built a streak of two hundred days, the prospect of losing it looms far larger than any positive reward, and the daily return becomes a defensive manoeuvre against loss rather than a pursuit of benefit. Duolingo's own reported figures make the mechanic's power plain: users who reach a seven-day streak are several times more likely to remain engaged, and the company built an entire secondary feature, the streak freeze, specifically to relieve the anxiety of at-risk users and reduce the rate at which they quit. Designers analysing the app have been candid that its success rests, in their telling, on design rather than on the superiority of its teaching, and that a large share of its revenue is organised around a single streak counter.

The third primitive is the notification, the interruption engineered to pull a user back into an application they had, however briefly, escaped. Notifications are tuned relentlessly. The phrasing, timing and frequency of a prompt are subjected to continuous experimentation, so that a change to a single word or a shift of an hour can be tested on a small cohort and rolled out to everyone if it lifts the numbers. This is not a caricature of how these products are built. It is how they are built, and it is described in these terms in the trade literature that celebrates them.

Set out plainly, the three mechanics form a loop: an unpredictable reward creates the pull, a streak raises the cost of leaving, and a notification supplies the nudge back in. What that loop optimises for is time on the product. That, and not any richer human outcome, is the thing being maximised.

Regulators have noticed. The United Kingdom's Age Appropriate Design Code, the statutory Children's Code enforced by the Information Commissioner's Office and in force since 2021, requires online services likely to be accessed by under-eighteens to avoid using nudge techniques that lead children into choices against their own interests, and asks providers instead to nudge towards wellbeing, including screen breaks. California modelled its own Age-Appropriate Design Code on the British one. The statute asked businesses to assess whether their products used system design features to increase, sustain or extend children's use, naming among them the automatic playing of media, rewards for time spent, and notifications, and it separately prohibited dark patterns that lead a child to take any action the business knows, or has reason to know, is materially detrimental to the child's physical health, mental health or well-being. Rewards for time spent. Notifications. The regulatory language is describing, feature for feature, the standard toolkit of engagement design, and it is describing it as a risk to be assessed and restrained when aimed at children. Turning that description into an enforceable rule has proved harder. Both provisions have been held up in the courts, and in March 2026 the Ninth Circuit Court of Appeals, in NetChoice v. Bonta, kept the dark-patterns prohibition blocked on the ground that it is likely void for vagueness. A business, the court reasoned, cannot know in advance what “materially detrimental” or “well-being” require of it when the range of possible harms runs from financial exploitation to lost sleep, distraction or hurt feelings. The regulators can name the mechanics. What they have not yet managed, at least in California, is to draw a line around them that a court will accept, and that difficulty matters for where this argument ends up.

How the same design crossed into the maths lesson

Here is the pivot on which the whole argument turns. Those mechanics did not stay on the social platforms. They crossed, wholesale, into educational software, because the companies building learning apps are drawn from the same talent pool, chase the same metrics and answer to the same kind of investor as the companies building everything else on a child's screen.

Walk through the feature list of a typical classroom learning product and you will find the loop reproduced faithfully. There are points and experience bars that fill towards a level. There are badges unlocked at intervals engineered to arrive just often enough to sustain interest. There are daily streaks with their attendant freezes and repair mechanics. There are leaderboards that convert solitary practice into social competition. There are, above all, notifications, and the design blogs that serve the sector discuss the timing and copy of those notifications with the same clinical attention to retention that you would find in any social app. The trade press for the industry openly frames gamification as a lever to boost engagement and retention, and the primary metric it reports is very often the time a child spends inside the product rather than anything the child has learned.

The defence offered for all this is that engagement is a precondition for learning. A bored child learns nothing, the argument runs, so a tool that holds attention creates the opportunity to teach. There is a grain of truth in it. Attention is necessary for learning. But necessary is not sufficient, and the sleight of hand in the industry's framing is to treat a necessary condition as though it were the outcome itself. A child can be intensely, measurably engaged by a maths app and be learning very little maths, in exactly the way a child can be intensely engaged by a slot-machine mechanic and be learning nothing at all. Engagement measures whether the child keeps tapping. It does not measure whether the child understands. These are different quantities, and the entire difficulty is that the first is trivially easy to measure and the second is genuinely hard.

This is the point Will Ellis of the campaign group Reclaim Childhood sharpened, in words Winkleman quoted from the ARC stage, arguing that engagement mechanics appear in learning apps “because they improve engagement, and engagement is what the business model requires,” not because they improve learning. The phrasing is worth dwelling on, because it identifies the exact place where the incentives fork. A subscription-funded or usage-funded educational product has a commercial interest in engagement that is direct, immediate and easy to measure in a dashboard. Its interest in learning is indirect, deferred and hard to measure at all, surfacing, if it surfaces, in a test score months later that no product analytics team will ever see. When two goals compete and only one of them shows up in the metrics that determine the next funding round, it is not cynical to predict which one wins. It is merely arithmetic.

Why engagement became the metric that ate the classroom

The deeper problem is that engagement optimisation and genuine pedagogy can actively pull in opposite directions, so that the better a product performs on the first, the worse it may be positioned to deliver the second. Consider the psychology of the external reward. A well-established body of research on motivation has repeatedly found that layering extrinsic rewards on top of an activity a child already finds interesting can reduce the child's intrinsic interest in that activity. Pay a child in points and badges for something they were happy to do for its own sake, and you risk teaching them that the thing is only worth doing for the points. For an entertainment product with no ambitions beyond retention, that trade is invisible and irrelevant. For an educational product whose entire justification rests on cultivating a durable disposition to read, to reason, to persist at hard problems, it is close to self-defeating.

There is a further mismatch in what the mechanics reward. Genuine learning frequently requires the opposite of a frictionless, dopamine-steady loop. It requires desirable difficulty, the productive struggle of retrieving a fact you have half-forgotten or wrestling with a problem just beyond your current reach. It requires tolerance of the boredom and frustration that precede mastery. It requires sustained attention held on a single hard thing for long enough to make progress, the precise faculty the Hertfordshire playground exposed as atrophied. Engagement design, by contrast, is built to eliminate friction, to keep the experience buoyant and rewarding and forward-moving, to ensure the child never sits with the discomfort long enough to be tempted to leave. The two philosophies are not merely different. At their extremes they are opposed. A product engineered so that a child never wants to stop is, almost by construction, a product engineered to remove the very resistances against which a mind grows stronger.

None of this proves that every gamified learning app harms every child. But it does establish that engagement and learning are separable, that they can conflict, and that the commercial machinery of the sector is tuned to the one that is easier to sell. That is enough to make the central question urgent rather than academic.

What the evidence actually shows about paper and screens

If the theoretical case is that engagement and learning diverge, the empirical case is that, in the aggregate and on average, screens have so far underdelivered on the outcomes that matter. The evidence is not unanimous and it is not always about engagement mechanics specifically, but it is substantial, and it has been accumulating for longer than this year's headlines suggest.

Begin with the medium itself. In a study published in the Oxford Review of Education in 2018, the University College London education researcher John Jerrim, working with John Micklewright and colleagues, took more than three thousand pupils across Germany, Ireland and Sweden and randomly assigned them to sit identical assessment items either on paper or on a computer screen. Randomisation matters here: because pupils were allocated by chance, the two groups were on average equivalent in ability, so any systematic gap in their scores could be attributed to the medium rather than to the children. The paper group scored roughly twenty points higher on the PISA scale, a gap the authors characterised as equivalent to about six months of additional schooling. The finding is part of a broader literature on what researchers call the mode effect, the tendency for the same knowledge to yield different measured performance depending on whether it is tested on paper or on a screen, an effect large enough in some countries to swing international rankings substantially. Jerrim's work is a caution about assessment as much as instruction. But its core implication is uncomfortable for the screen-first classroom: the digital medium was not neutral. It cost pupils something.

Then consider what happens in the brain during the physical act of learning. Audrey van der Meer, a neuroscientist at the Norwegian University of Science and Technology, has spent two decades comparing the neural signatures of handwriting and typing. In a high-density electroencephalography study published in Frontiers in Psychology, her team fitted participants with nets of well over two hundred sensors and recorded their brain activity as they were prompted, repeatedly, either to write a word by hand or to type it. Writing by hand produced markedly more elaborate patterns of connectivity across the brain, with widespread coordination between regions in frequency bands that van der Meer and her collaborators associate with memory formation and the encoding of new information. Typing did not. Her interpretation is that the precisely controlled movements of forming a letter by hand feed a stream of visual and motor information into the brain that helps knit together the connections underlying learning, and that this stream is largely absent when the same word is produced by tapping identical keys. The sample sizes in this kind of laboratory work are small, and connectivity is not the same thing as a test score, so the finding should be held with appropriate care. But it converges with the behavioural evidence rather than contradicting it, and the direction of travel is consistent: the embodied, effortful, analogue act engages more of the learning brain than its frictionless digital substitute.

Recent reviews of the research literature frame the same tension in their own terms. A scoping review by Urška Žnidarič, Erik Štrumbelj and Octavian Machidon, synthesising 937 papers on the negative effects of digital technology on cognition, found that the harms it could identify ran largely through displacement, the crowding out of other activities, rather than through technology being harmful in itself, and that effects were often moderated by socioeconomic and environmental factors. Its organising conclusion is what the authors call an efficiency-atrophy paradox: digital tools can improve short-term performance while potentially weakening long-term, unassisted cognition. That is the engagement-versus-learning problem restated in cognitive terms. A tool can make today's task go better and leave the mind less able to do it alone tomorrow, and a dashboard measuring today's performance will record only the gain. A second review, a systematisation of knowledge by Supriya Khadka and Sanchari Das covering 111 studies of extended reality in the education of children aged three to eight, found that pedagogy received by far the most scholarly attention, while data security practices received almost none, and children's health and accessibility for disabled children were also thinly covered. Even for children as young as three, the research effort has gone overwhelmingly into whether a technology can teach, and comparatively little into what else it does to the child, and to the child's data, while it is teaching.

That deterioration is not merely theoretical either. It is legible in the international data that governments themselves collect. Sweden's national performance on the PISA assessments fell over the last decade, and in 2022 almost a quarter of fifteen and sixteen-year-olds, 24 per cent, did not reach a basic level of reading comprehension. It was against exactly this backdrop that the country began reconsidering the enthusiasm with which it had digitised its classrooms. Correlation is not cause, and the OECD's education director, Andreas Schleicher, has urged caution about attributing cause and effect; reading declines have many parents. But when a country that went further and faster than almost anyone into one-to-one devices sees its literacy soften and responds by buying paper textbooks again, the sequence is worth taking seriously.

The pushback gathers institutional weight

For years the case against screen-saturated schooling was made largely by individual sceptics, and it was easy to wave away as nostalgia. What changed this year is that the argument acquired institutions, budgets and votes.

The most consequential move came in Los Angeles. The board of the Los Angeles Unified School District, the second-largest public school system in the United States, voted unanimously to create and enforce district-wide limits on classroom screen time, the most sweeping such reduction adopted by any major American district, and the policy took effect when the 2026-27 school year opened in August. It is phased. Since August, children from preschool to first grade have had no device time at all, outside narrow exceptions such as mandated assessments and support required for pupils with disabilities. From November, second and third graders are capped at twenty minutes a day and fourth and fifth graders at thirty. Secondary schools follow in January 2027 with weekly caps per subject, an hour a week in each subject in middle school and an hour and a half in high school. Devices no longer go home by default; families must opt in. Student access to YouTube, social media and streaming platforms is blocked unless a teacher approves it. Behind the resolution was Schools Beyond Screens, a national coalition founded by district parents and teachers, whose campaigning the board member Nick Melvoin credited during the meeting. The significance is less in the specific minute-by-minute limits, which are debatable and will be contested, than in the precedent: a flagship district conceded that unrestricted classroom screen time was a problem serious enough to legislate against, rather than a self-evident good to be expanded.

Los Angeles is not alone, and the American movement is broader than any single district. At least six states approved screen-time legislation in 2026. Iowa capped digital instruction from kindergarten to fifth grade at sixty minutes a day, with exceptions for special education, computer science and assessments. Utah directed its state board of education to write grade-specific technology rules, including screen-time limits for the earliest grades and an end to one-device-per-pupil ratios in elementary school. Tennessee requires every district serving kindergarten to fifth grade to adopt a policy minimising unnecessary screen time, and Alabama has set screen-time standards for childcare, prekindergarten and kindergarten. Virginia and Maine went more gently, the first requiring schools to teach pupils about device use and its addictive potential, the second commissioning a study of how much instructional time is spent on school-issued devices. Not every bill succeeded: measures in Missouri and Oklahoma passed their state houses and stalled in the senate. In New York, Governor Kathy Hochul has said she is open to limiting screen time for younger pupils, citing concern from parents and educators about the youngest children spending their days on individual laptops and tablets, but has proposed no statewide limit. New York City moved first, announcing in September 2026 a one-year ban on student-facing generative AI from preschool through eighth grade, restrictions on individual devices for the youngest children, and recommended daily caps of thirty minutes for grades three to five and forty-five for middle school. The precise legal force varies from place to place and should not be overstated, but the pattern is unmistakable: a coordinated, bipartisan reappraisal of a decade of classroom digitisation.

Europe has moved in parallel and, in places, further. The regional government of Madrid announced restrictions phasing out individual digital devices across publicly funded early-childhood and primary schools, affecting well over half a million pupils, and pushing one-to-one tablets and laptops out of the education of children up to the age of twelve in favour of shared use under a teacher's supervision, with narrow exceptions for special educational needs and subjects such as coding that inherently require the tools. Sweden, having been among the most aggressively digitised school systems in the world, reversed course under the slogan “från skärm till pärm”, from screen to binder, and by April 2026 had allocated schools more than 2.1 billion kronor in grants for printed textbooks and teacher guides, with a new textbook-based curriculum due in 2028, an explicit response to slipping results. Denmark has earmarked DKK 540 million over the coming decade to put printed books and physical teaching materials back into its classrooms. Taken together, these European commitments run into the hundreds of millions of euros, spent not to add technology but to remove it. That is a remarkable thing for finance ministries to fund, and it signals that the reappraisal has moved from opinion pages into public budgets. Britain, where this essay began, is at an earlier stage. In June 2026 the Department for Education opened a call for evidence on screen use by children aged five to sixteen, with a section devoted specifically to screens in schools, to inform future policy.

The counter-case for the well-designed screen

Honesty requires taking the strongest version of the opposing argument seriously, because it is real and it is not trivial, and a precautionary case that ignores it deserves to lose.

Educational technology, at its best, does things that paper and a single overstretched teacher cannot. For a child with dyslexia, text-to-speech and adjustable typography can be the difference between access to the curriculum and exclusion from it. For a blind child, a screen reader is not a distraction machine but a door. For a child with a physical disability that makes handwriting painful or impossible, a keyboard is a liberation, and the neuroscience celebrating the handwriting brain must not be turned into a cudgel against children for whom the pen was never a neutral option. Assistive technology is one of the genuine triumphs of the digital classroom, and any policy that sweeps it away in a general panic about screens will harm precisely the children with the least margin to spare.

Well-designed software also personalises in ways a class of thirty makes impossible. Adaptive systems that adjust the difficulty of practice to the individual child, that give immediate feedback on a wrong answer rather than making a pupil wait a week for a marked worksheet, that let a fast child accelerate and a struggling child consolidate without public embarrassment, can genuinely differentiate instruction. There is a reasonable evidence base that some carefully built, curriculum-aligned programs, used in moderation for specific purposes, produce real gains, particularly in structured domains such as early numeracy and foundational reading. In under-resourced schools, a device can carry a library, a set of translated materials for a child new to the language of instruction, and a bank of practice a school could never afford to print. To pretend none of this is valuable is to lose the argument on the facts.

And not every engagement mechanic is a slot machine. A progress bar that shows a child how far they have come through a genuinely well-sequenced curriculum is providing meaningful feedback, not manufacturing a compulsion. Immediate confirmation of a correct answer is sound pedagogy with a century of evidence behind it. Some gamification supports learning rather than parasitising it, and the honest position is not that gamified equals harmful but that the category contains both, and that the two are hard to tell apart from the outside. Which is precisely the problem, and precisely why the argument has to move from banning features to allocating proof.

Where the burden of proof belongs

Strip the debate down and it reduces to a single question of institutional design: when a school is deciding whether to install a piece of software on the devices of hundreds of children, who has to demonstrate what, and to whom, before it goes in?

Today the answer, overwhelmingly, is that nobody has to demonstrate very much. A product arrives with impressive engagement statistics, a glossy pilot, testimonials, perhaps a thin white paper of its own commissioning, and the default is deployment. The burden, such as it is, falls on the sceptic. A worried teacher, a dissenting parent, a cautious administrator must assemble the case against, must find the evidence of harm, must argue upstream against a purchase already made and a contract already signed. The product is presumed beneficial until someone can prove it is not, and because proving harm in a diffuse population of children over the timescales on which learning and attention actually develop is extraordinarily difficult, the presumption is nearly impossible to dislodge. The structure of the burden guarantees the outcome.

The argument of this essay is that the burden is on the wrong shoulders, and that the reason it is on the wrong shoulders is that we have quietly classified educational software as a consumer product rather than as what it functionally is, an intervention imposed on a captive population of children who cannot consent, cannot leave, and cannot tell whether the thing holding their attention is teaching them or merely holding their attention. We do not let food manufacturers put a substance in school lunches on the strength of the fact that children keep eating it. We do not let a pharmaceutical company deploy a treatment across a paediatric population because early users engaged with it enthusiastically. In domains where the subjects are children and the stakes are developmental, we place the burden of proof on the party introducing the intervention, and we ask them to demonstrate benefit, and the absence of harm, before deployment rather than after.

Applied to the classroom, burden-shifting has a concrete shape, and it is more modest than the word precaution sometimes implies. It means that a company selling a product as educational should be required to show that the product produces learning, measured as learning and not as engagement, before it is bought at scale, and to disclose the mechanics it uses to sustain attention so that a purchaser can weigh them. It means that the primary metric a vendor is permitted to advertise to a school should be an outcome, not a retention statistic, because a retention statistic sold as if it were an educational result is close to a category error. It means that a regulator who has already identified, in the Children's Code and in California's attempt to follow it, rewards for time spent and engineered notifications as risks when a social platform aims them at a child, should ask why the identical mechanic earns a pedagogical exemption the instant the same child meets it inside a maths app. If a design pattern is a dark pattern in one context, the context does not launder it. The child's developing brain does not know or care which app store category the software was filed under. California's experience in court is also an argument for this approach over a ban. A prohibition must define in advance which features are “materially detrimental” to a child's “well-being”, and that is exactly where the Ninth Circuit found it too vague to enforce. A requirement that the vendor show its product produces learning asks a narrower question, one that can be answered product by product, by the party holding the data.

Washington has, somewhat unexpectedly, conceded the principle. On 20 August 2026 the US Department of Education issued a Dear Colleague letter, signed by Kirsten Baesler, the assistant secretary for elementary and secondary education, which K-12 Dive read as the department going to the industry's defence against the screen-time backlash. Its argument was that educational technology should not be judged mainly by how long children spend in front of it, and that recreational and educational technology should not be treated as the same thing. Yet the yardstick it offered in place of minutes is the one this essay has been arguing for. Investments, the letter said, “should be guided by evidence of effectiveness”; states making procurement and renewal decisions should look for products that have run randomised controlled trials and demonstrated positive outcomes; and providers should publish rigorous, independent evaluations of their impact on student learning “whenever feasible”. That is a federal agency endorsing outcome evidence over time on device. It is also, in its own phrasing, a picture of the gap. A letter of guidance binds no one, the decisions remain with states and districts, and “whenever feasible” leaves the vendor to judge when feasibility applies. The market is already reading the signal. In July, Ed Corey, chair of the Torrington school board, told EdWeek Market Brief that companies whose products rely heavily on daily screen time will need to show a “legitimate curriculum use case” backed by research, and a national survey of 600 teachers in May found that 85 per cent believed pupils spend too much class time on screens. The demand for proof is arriving. What has not arrived is any rule that says it must be met before the product goes in.

There is an obvious objection, and it deserves a straight answer. Proving educational benefit is hard, slow and expensive, and a strict evidence bar would slow the sector, entrench incumbents who can afford large trials, and deny some genuinely good tools to the schools that most need them. This is true, and it is the strongest thing that can be said against burden-shifting. But notice that the difficulty of proving benefit is not a neutral fact about the world. It is the very thing that has allowed engagement to stand in for learning for a decade, precisely because engagement is easy to prove and learning is hard, and a regime that rewards whatever is easy to measure will get more of what is easy to measure and less of what matters. The answer to a hard measurement problem is not to abandon the measurement and default to deployment. It is to make the party with the commercial interest, and the resources, and the data, carry the cost of the proof, calibrated sensibly to the risk, with lighter obligations on assistive tools whose benefit is manifest and heavier ones on attention-maximising products aimed at the youngest children. Burden-shifting does not mean banning software. It means the person who profits from the claim has to substantiate the claim.

The precautionary turn and what it asks of everyone

The medicine analogy is imperfect, and it should not be pushed to the point of demanding randomised controlled trials for every worksheet generator. But it captures the essential asymmetry the current arrangement has inverted. When you intervene in the development of a child who cannot consent and cannot exit, and when the potential harm is to faculties, attention, imagination, the capacity for effortful thought, whose damage may not surface for years, you do not get to run the experiment live on the whole population and check the results later. The cost of being wrong is borne by children, irreversibly and at the precise moment their minds are being formed, while the benefit of being right accrues to a firm in the form of engagement it can bank this quarter. That asymmetry is the entire moral case for shifting the burden, and it does not depend on the alarmism of any single conference talk to hold.

Each party in the system has an obligation that follows from this. The companies owe honesty about what their products optimise for, disclosure of the engagement mechanics they embed, and evidence of learning rather than evidence of stickiness, and the better firms should welcome a standard that separates them from competitors selling compulsion in a curricular costume. Regulators owe consistency, and the simplest reform available to them is to stop pretending that a mechanic they have identified as harmful becomes benign on crossing from a social app into an educational one, and to require that products marketed to children as teaching tools clear a bar proportionate to that claim. And schools, which are the point of deployment and therefore the point of leverage, owe their pupils the institutional courage to demand outcomes rather than engagement dashboards, to run their own small comparisons, to keep the paper and the pen and the unstructured hour that the market has no incentive to defend, and to treat any vendor unwilling to be measured on learning as having answered the question by declining to.

The headteacher in Hertfordshire did something the market will never do on its own. She created a condition, an empty playground, an hour without stimulation, in which the absence of the product became visible, and with it the shape of what the product had been quietly displacing. The children's inability to fill that hour is not proof of anything on its own; it is one anecdote, and anecdotes are where inquiries begin, not where they end. But it names the thing at stake with a precision the engagement statistics never will. The question is not whether children are looking at screens. It is whether the screens are teaching them, or merely keeping them, and whether we are willing to make the people who profit from the difference be the ones who prove it. Until that burden moves, the honest description of much of what is installed on the classroom device is not an educational tool that happens to be engaging. It is an engagement machine that happens to be pointed at a child in a school, and the fact that it is pointed at a child in a school is exactly why the proof should come first.

References

  1. Winkleman, Sophie. “Why AI in the Classroom is a Catastrophe.” Talk delivered at the Alliance for Responsible Citizenship (ARC) 2026 conference, Olympia, London, June 2026. Video, YouTube (Alliance for Responsible Citizenship channel): ; transcript, The Singju Post: https://singjupost.com/transcript-why-ai-in-the-classroom-is-a-catastrophe-sophie-winkleman/
  2. Jerrim, John, John Micklewright, Jörg-Henrik Heine, Christine Salzer and Caroline McKeown. “PISA 2015: how big is the 'mode effect' and what has been done about it?” Oxford Review of Education, vol. 44, no. 4, 2018. https://www.tandfonline.com/doi/abs/10.1080/03054985.2018.1430025
  3. Van der Meer, Audrey L. H., and F. R. (Ruud) van der Weel et al. “Handwriting but not typewriting leads to widespread brain connectivity: a high-density EEG study with implications for the classroom.” Frontiers in Psychology, vol. 14, 2024. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1219945/full
  4. Žnidarič, Urška, Erik Štrumbelj and Octavian Machidon. “A Scoping Review of the Negative Effects of Digital Technology on Cognition.” arXiv preprint 2603.10025 [cs.CY], v1 23 February 2026, revised v2 15 August 2026. https://arxiv.org/abs/2603.10025
  5. Khadka, Supriya, and Sanchari Das. “SoK: Understanding the Pedagogical, Health, Ethical, and Privacy Challenges of Extended Reality in Early Childhood Education.” Augmented Humans 2026; arXiv preprint 2602.12749 [cs.HC], 13 February 2026. https://arxiv.org/abs/2602.12749
  6. BBC News. “Back to books: Sweden's schools give up digital learning.” 16 April 2026. Republished by The Star (Kenya): https://www.the-star.co.ke/news/world/2026-04-16-back-to-books-swedens-schools-give-up-digital-learning
  7. Cohen, Joshua. “Why Swedish Schools Are Bringing Back Books.” Undark, 1 April 2026. https://undark.org/2026/04/01/sweden-schools-books/
  8. EdSource. “Los Angeles Unified to limit screen time for all students, prohibit use among youngest students.” 2026. https://edsource.org/2026/classroom-screen-time-limits-policy/756523
  9. K-12 Dive. “Screen time limits coming for LAUSD in the fall.” 2026. https://www.k12dive.com/news/lausd-imposes-screen-time-limits-starting-in-2026-27/818224/
  10. Dale, Mariana, and Elly Yu. “LAUSD's youngest students will start a screen-free school year. Here's what parents need to know.” LAist, 30 July 2026. https://laist.com/news/education/lausds-youngest-students-will-start-a-screen-free-school-year-heres-what-parents-need-to-know
  11. Patino, Vania. “In a new chapter for classroom learning, LAUSD is welcoming students back with screen time limits.” Spectrum News, 3 August 2026. https://spectrumlocalnews.com/ca/california/education/2026/08/03/screen-time-limits-lausd
  12. Stateline. “States move to limit classroom screen time.” 23 July 2026. https://stateline.org/2026/07/23/states-move-to-limit-classroom-screen-time/
  13. Rauf, David Saleh. “The Next Challenge for K-12 Ed-Tech: Proving Classroom Screen Time Is Worth It.” EdWeek Market Brief, 10 July 2026. https://marketbrief.edweek.org/regulation-policy/the-next-challenge-for-k-12-ed-tech-proving-classroom-screen-time-is-worth-it/2026/07
  14. Gothamist. “Gov. Hochul says she's open to limiting screen time for younger students in NY schools.” https://gothamist.com/news/gov-hochul-says-shes-open-to-limiting-screen-time-for-younger-students-in-ny-schools
  15. Ginsburg, Aaron. “NYC announces one-year ban on generative AI for public school students through 8th grade.” 6sqft, 2 September 2026. https://www.6sqft.com/nyc-ai-ban-public-schools/
  16. Aleteia. “Madrid schools limit tech use for half a million students.” 4 August 2025. https://aleteia.org/2025/08/04/madrid-schools-limit-tech-use-for-half-a-million-students/
  17. The Danish Dream. “Danish Schools Reintroduce Textbooks to Improve Literacy.” https://thedanishdream.com/education-in-denmark/danish-schools-reintroduce-textbooks-to-improve-literacy/
  18. Information Commissioner's Office (ICO). “Age appropriate design: a code of practice for online services (the Children's Code),” standard 13, Nudge techniques. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/age-appropriate-design-a-code-of-practice-for-online-services/13-nudge-techniques/
  19. Department for Education. “Screen use by children aged 5-16: call for evidence.” Opened 8 June 2026, closed 29 June 2026. https://consult.education.gov.uk/screen-time-policy-team/screen-use-by-children-aged-5-to-16-cal/
  20. United States Court of Appeals for the Ninth Circuit. NetChoice, LLC v. Bonta, No. 25-2366, opinion filed 12 March 2026. https://caselaw.findlaw.com/court/us-9th-circuit/118216051.html
  21. Cooley LLP. “NetChoice v. Bonta: Ninth Circuit Narrows Injunction Against California's Age-Appropriate Design Code Act.” 30 March 2026. https://www.cooley.com/news/insight/2026/2026-03-30-netchoice-v-bonta-ninth-circuit-narrows-injunction-against-californias-ageappropriate-design-code-act
  22. Future of Privacy Forum. “An Analysis of the California Age-Appropriate Design Code.” October 2022. https://fpf.org/wp-content/uploads/2022/10/FPF-Policy-Brief-California-Age-Approp-Design-Code-R2.pdf
  23. Quartr. “Keeping the Streak Alive: The Story of Duolingo.” https://quartr.com/insights/edge/keeping-the-streak-alive-the-story-of-duolingo
  24. THE Journal. “Department of Education Issues New Guidance on Education Technology and Screen Time.” 20 August 2026 (reporting the Dear Colleague letter signed by Kirsten Baesler, Assistant Secretary, Office of Elementary and Secondary Education). https://thejournal.com/articles/2026/08/20/department-of-education-issues-new-guidance-on-education-technology-and-screen-time.aspx
  25. Merod, Anna. “Education Department goes to ed tech's defense in new guidance.” K-12 Dive, 20 August 2026. https://www.k12dive.com/news/education-department-goes-to-ed-techs-defense-in-new-guidance/828425/

Tim Green

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Support What We Do

Discuss...