The Fluency Trap

Generation Z is the first generation not to surpass its parents in knowledge and ability, argues the French neuroscientist Michel Desmurget in his book La fabrique du crétin digital. Hungarian pupils' PISA scores, too, have been falling for a decade. The problem is subtler than laziness: the fluency illusion, the effortlessness ChatGPT offers, and the way digital tools remove much of the effort from learning — when effort is the secret ingredient. This part examines the collapse of education and the new threat of emotional attachment to AI.

1. The PISA nosedive — Hungarian and global data

The Programme for International Student Assessment (PISA) — the OECD's international survey of 15-year-olds, run every three years — is the most reliable gauge of how a country's education system performs relative to its international peers. The most recent round, in 2022, makes uncomfortable reading for Hungary: in mathematics its pupils scored 473 points, level with the OECD average of 472, but well below the peak of 491 recorded in 2006. Reading produced 473 points (OECD average: 476) and science 486 points (OECD average: 485). Hungarian pupils, in other words, perform around the OECD average — but over the past decade and a half, they have steadily lost ground.

The global picture points the same way. In La fabrique du crétin digital, the French neuroscientist Michel Desmurget makes a startling claim: over the past 100 to 120 years, every generation has surpassed its parents in ability and knowledge. Generation Z is the first not to — the first, in his account, to be falling behind rather than advancing, the first to fall short, the first not to outperform its parents. This "generational reversal" is not merely a statistical anomaly; it is one of the most measurable social consequences of what the digital environment does to the brain.

Several factors are contributing to the decline in PISA results. Pandemic learning loss explains part of it — the 2022 round was the first big international comparison after Covid — but the trend is not new. Hungarian mathematics scores have been on a downward path since 2009, and reading skills have been eroding steadily too. So what else is at work, beyond the pandemic? The causes include both the spread of digital technology and structural problems in the education system: teacher shortages, funding difficulties and an excessive emphasis on digital tools all disrupt how learning actually takes place.

2. The fluency illusion — the trap of ease

The most important effect of digital technology on education is not its distracting presence but a far subtler neuropsychological phenomenon: the fluency illusion. It works like this: when something is easy to do, we take that ease as evidence that we know it far better than we actually do. Research finds that pupils for whom studying felt "easy" — those who reread a text several times without testing themselves — performed far worse in exams than those for whom it felt "hard": a single read-through, followed immediately by an attempt to recall what they had taken in.

Digital tools — from the pocket calculator to ChatGPT — offer precisely that ease. When a pupil asks ChatGPT for an essay, there is no need to think through the structure, hunt down the arguments or put the ideas into words. The model does the work for them. The result: the pupil feels they understand the subject, when in fact they have only skimmed the surface. Research published by the Stanford Digital Economy Lab in August 2025 (Brynjolfsson, Chandar and Chen) found that since ChatGPT arrived, employment among workers aged 22 to 25 in the occupations most exposed to AI has fallen by roughly 13% in relative terms — and the decline is sharpest precisely among those at entry level, straight out of education.

The biology of learning shows that effort is not the enemy but the secret ingredient. When the brain has to work hard to process information, stronger memory traces form. This is the principle of desirable difficulty, described by the cognitive psychologist Robert Bjork. Digital tools do away with that difficulty, and in doing so they weaken the depth of learning.

The fluency illusion shapes more than school results — it shapes the capacity for critical thinking. When a pupil receives a finished answer from ChatGPT, what they skip is not only the writing but the internal debate involved in reasoning. Reasoning is not a linear process: you formulate a thought, then question it, revise it, discard it, think it through again. That "cognitive friction" — the tension between competing arguments — is what builds deep understanding. Generative AI removes that friction and produces a smooth, shallow, plausible-sounding answer instead. A much-discussed 2025 experiment at the MIT Media Lab (Kosmyna and colleagues, "Your Brain on ChatGPT") made this measurable: participants were split into three groups — brain-only, search engine and ChatGPT — and their brain activity was recorded by EEG while they wrote essays. The ChatGPT users' brain networks showed the weakest activity, and by the end of the experiment, they remembered less of the content of the essays they had themselves produced with AI. The researchers called this cognitive debt: the easier route comes at a cost, and the learning loss simply becomes apparent later.

The scale of the phenomenon is troubling in itself. A 2024 survey by Common Sense Media found that 70% of American teenagers had used a generative AI tool at some point, many of them for homework — often taking the answer they were given without checking it. This is not "asking for help"; it amounts to replacing the learning process altogether. The consequence: a generation that can produce an "acceptable" answer quickly, but struggles when no ready-made solution exists.

3. The multitasking neuro-myth and the fragmentation of attention

Generation Z pupils often take pride in saying that they can multitask — following an online lesson while chatting and scrolling TikTok. Neuroscience flatly contradicts this: the human brain cannot multitask. What actually happens is "task-switching": the brain flips rapidly from one task to another, but never processes them in parallel.

Task-switching imposes a substantial cognitive cost. The classic research by Meyer and Rubinstein found that every switch consumes time and cognitive capacity. When a pupil glances at their phone during an online lesson, the brain has to "unload" the content of the lesson, "load" the message on the phone, then "reload" the lesson. The process does not merely slow things down; it reduces the quality: instead of deeper processing, the result is shallow, fragmented memory traces.

In digital education, this problem is especially acute. PowerPoint presentations, PDFs and online exercises often push pupils into passive consumption — they read, watch and listen, but never actively work through the material. The fluency illusion is at its most potent in this setting: the pupil feels they have "covered" the material because they sat through the presentation, when in truth they have skated across the surface.

The consequences of fragmented attention reach beyond school results. In his book Deep Work, the Georgetown computer scientist Cal Newport sets out in detail why deep, concentrated attention is a precondition not only of learning but of creativity and problem-solving. Deep work — the ability to concentrate on a complex task for long stretches without distraction — is a mental state in which the brain processes information most efficiently. Yet the generation growing up in a digital environment experiences that state less and less often: phone notifications, the pull of social media and the culture of multitasking together shape a pattern of "shallow work", in which the brain is always busy but never goes deep.

Task-switching impairs more than school results; it erodes social skills too. When a pupil is looking at their phone during a lesson, what they miss is not only the material but the interaction with the teacher and their classmates. Eye contact, non-verbal cues, shared laughter, arguments — these are social experiences that communication through a screen cannot replace. On "Digital Predators", an episode of the Hungarian public-television talk show Ridikül, the family-safety specialist Ákos Pöltl issued a warning: young people are physically present at school but mentally absorbed in the online world — and constant screen use also erodes their ability to concentrate. This gap between physical and mental presence undermines both school performance and the quality of social relationships.

None of this rules out training attention, however. Research indicates that attention is a skill that can be practised, much like a muscle. Mindfulness exercises, digital-detox experiments and the introduction of phone-free zones in schools are all measures that can help young people regain their ability to concentrate. The question is not whether it is possible, but whether the current structure of education supports it — or further entrenches the fragmentation of attention.

4. AI as a substitute for human attachment — beyond education

The effects of ChatGPT and generative AI on education go beyond the fluency illusion. Another emerging and still barely researched phenomenon is the use of AI as a substitute for human attachment — young people using artificial intelligence not merely as a study aid but as a stand-in for a social relationship.

Character.AI and other AI companion apps have 20 million monthly users, most of them under 24. These applications let young people "talk" to virtual characters — whether historical figures or invented companions. The psychological risk is that the sense of "empathy" and "care" an AI can evoke may activate emotional feedback loops in the brain that, until now, only human relationships could trigger.

The American Psychological Association's June 2025 health advisory names the erosion of real-world relationships as one of the central risks of AI tools for the young, and calls for specific safeguards for adolescent users. The problem is not that young people talk to an AI — it is that the conversation becomes a substitute for human relationships. Where the AI is always understanding, always available and never critical, real human relationships — which come with conflict, disappointment and compromise — grow steadily less appealing.

5. Counter-arguments and qualifications

The critique of digital education is itself open to challenge. First, technology is not inherently bad — the problem is how it is used. Interactive simulations, digital libraries and the possibilities of distance learning are all tools that, within the right pedagogical framework, can enrich education.

Second, the "Generation Z has failed" narrative is too simplistic. The fall in PISA scores also reflects the pandemic, the crisis in teacher recruitment, funding problems and widening social inequality. Digital technology is only one factor among many — and not necessarily the most important.

Third, "traditional" education was hardly perfect. Chalk-and-talk lecturing from the front, assessment built on factual recall and the absence of individual differentiation have been problems in Hungarian schooling for a long time. If nothing else, the arrival of digital technology has thrown those structural problems into relief.

6. Consequences — what to watch for

The consequences of the collapse in education bear directly on Generation Z's prospects in the labour market. If young people do not acquire critical thinking, complex problem-solving and the ability to learn deeply at school, they start at a disadvantage in a labour market where those skills are crucial. And at just this point — as generative AI begins to squeeze out entry-level work — that disadvantage may prove especially painful.

These are the indicators to watch:

The trend in PISA scores. The results of the 2025 PISA round, expected to be published in 2026, will show whether the post-pandemic decline was temporary or whether it has become entrenched. In Hungary, the mathematics results will be particularly worth watching, since that domain correlates most strongly with later success in the labour market.

The regulation of ChatGPT and similar tools in schools. A growing number of countries and schools are restricting the use of generative AI in exams and homework. The question is whether such restrictions help restore effort-based learning, or merely treat the symptom.

The rise of emotional attachment to AI. The growing popularity of Character.AI and comparable platforms, particularly among younger users, points to a new form of digital dependency — one that education policy and mental health services are not yet adequately addressing.

In sum: the collapse of education is not Generation Z's fault — it is the result of a system that has prioritised technological convenience over deep learning. The fluency illusion, the myth of multitasking and attachment to AI may together shape a generation adept at handling surface-level information but ill-equipped for complex, unstructured problems. And those are precisely the problems waiting for them in the labour market — where generative AI is already eliminating entry-level roles. The next instalment in this series examines the consequences of that combination, in an analysis of the labour market.

Over the past 100 to 120 years, every generation has surpassed its parents in ability and knowledge. Generation Z is the first not to.

— reported argument, Michel Desmurget (2019), La fabrique du crétin digital

Using ChatGPT to learn is like taking a helicopter to the summit: the view is just as good, but you never learn to climb. At the next mountain you are on your own.

— Pedagogical observation on the fluency illusion