The real gap of the future isn't AI: it's critical thinking. Artificial intelligence training is needed for businesses, schools, and families.
- Sergio Bonuomo
- 2 days ago
- 7 min read
The true inequality of 2026 will not be measured by the ability to use Generative AI, but by the ability to thinkbefore questioning it, to evaluate its results, and to sustain sustained cognitive effortin a world that rewards scrolling. The 2022 OECD-PISA data, the election of brain rot as Oxford's word of the year as early as 2024, and the European artificial intelligence literacy requirement that came into force on February 2, 2025, all reveal the same phenomenon: AI doesn't cause the decline in cognitive skills; it amplifies it and makes it impossible to hide.
Three data that measure a real phenomenon
The latest numbers leave no room for optimistic interpretations.
The 2022 OECD-PISA report found a decline of fifteen points in mathematics in Italy compared to 2018 , in line with the average decline in OECD countries. Reading scores held up at the aggregate level, but high schools are showing a downward trend, and the gap between students in Northern and Southern Italy remains between fifty and seventy points, a gap equivalent to about a year and a half of schooling.
In December 2024, Oxford University Press named brain rot its word of the year. The term's frequency of use increased by 230 percent between 2023 and 2024. The official definition is the alleged deterioration of a person's mental or intellectual state , believed to be the result of excessive consumption of trivial or unstimulating online material. It's not a recent invention: the expression first appeared in Henry David Thoreau's Walden in 1854. It was Generation Z and Generation Alpha (the same ones exposed to the phenomenon) who brought it back into circulation.
The Italian Publishers Association, in data published in December 2025, confirms another erosion: the average weekly time spent reading print books in Italy dropped from three hours and thirty-two minutes in 2022 to three hours and seven minutes in 2025. The number of print readers is growing (we've reached seventy-six percent of the population), but the actual time spent reading is decreasing. The more people read, the less time they spend doing so.
Why brain rot is not a lexical fad
The Oxford election isn't linguistic folklore. It's a diagnosis the culture has given itself. Thoreau used brain-rot to criticize society's tendency to devalue complex ideas in favor of simple ones. One hundred and seventy years later, the expression is being taken up by generations most exposed to digital hyperstimulation to describe what they feel is happening to themselves.
The interesting cultural fact is precisely this: self-diagnosis . Those who talk about brain rot aren't passively suffering it; they're aware of it. Yet it persists. This signals that the problem isn't informational (we all know that eight hours of scrolling isn't good for you) but structural: the attentional architecture has been reprogrammed from the outside.
The three cognitive erosions that are taking hold
Talking about a decline in critical thinking is a generalization. It's worth breaking down the phenomenon into three distinct erosions that educators, trainers, and managers are consistently observing.
The first is sustained attention. The ability to remain on a non-rewarding task for forty, sixty, or ninety minutes straight. This isn't clinical hyperactivity: it's a neurological threshold accustomed to receiving a reward every eight to fifteen seconds. When the reward doesn't arrive, the system looks elsewhere.
The second is deep reading. Reading a three-thousand-word article, reformulating it, identifying the implicit thesis, and recognizing the unstated premises. The average time spent on the page of major online publications has halved in the last five years. Reading has become scanning.
The third is tolerance for boredom and uncertainty . Creativity and complex problem solving require inhabiting the void—not yet knowing the answer, not yet having the solution. A generation raised to expect every doubt to be resolved in three seconds with a prompt has never had to exercise that muscle.
The Perspective Error: AI Is Not the Cause, It's the Detector
Here comes the point that those who work with artificial intelligence know every day, but which continues to elude public debate. Generative AI didn't create cognitive deficits. It inherits them. It amplifies them. It makes them impossible to hide.
A professional who can't formulate a problem won't be able to write a useful prompt . They'll receive a plausible answer and accept it, because they don't have the tools to question it. A manager who can't read a contract in depth won't recognize that the AI-generated summary has omitted the crucial clause. A student who has never tolerated the tedium of rereading a difficult text three times won't understand why the answer from ChatGPT, Claude, Gemini, etc., however fluent, isn't an argument but an imitation of one.
Artificial intelligence is a mirror. It reflects exactly the quality of thought you bring to it. The crisis isn't AI's: it's the thinker's.
What's changing with the AI Act: Literacy is now mandatory
On 2 February 2025, Article 4 of Regulation (EU) 2024/1689, known as the AI Act, entered into force. It imposes on providers and users of artificial intelligence systems the obligation to ensure a sufficient level of AI literacy among their staff .
The European Commission clarified in its FAQs of May 13, 2025, that the obligation also applies to companies whose staff use generative tools for writing or translation. Italy, the first Member State, enacted Law 132/2025 to implement the law, designating the National Cybersecurity Agency as the competent authority.
The regulatory definition of AI literacy should be read carefully: it includes the ability to recognize risks, opportunities, and potential impacts. In other words, AI literacy isn't knowing how to start a chat with Claude. It's knowing when it's wrong.
This transforms critical thinking from a soft virtue to a compliance requirement . Companies that limit themselves to two-hour technical courses on how to use the prompt will not be compliant. Those that teach how to verify the sources AI cites, how to recognize hallucinations, how to distinguish correlation from causation in summarized data, and how to challenge the output will be.
Artificial intelligence training is needed for businesses, schools, and families. Three concrete steps for those who want to act now.
The solutions proposed usually stop at reading more books to children. True, but insufficient. Here's a more practical map, broken down into three target audiences.
For parents
Three measurable interventions. First: introduce screen-free time slots, especially the first hour of the morning and the last hour of the evening, when learning is consolidated. Second: replace—not ban—scrolling with activities that require sustained attention , such as board games, shared reading, or building blocks. Third: model the behavior , because a parent who scrolls in front of their child invalidates any prohibition.
For school
The 2025 National Guidelines introduce Computer Science as a cross-curricular core and explicitly mention algorithmic literacy . There are three directions of action. First: design assessments that require extended reasoning , not short answers that an LLM student could recite in a second. Second: use AI as an object of critical analysis in the classroom —what it did wrong, why, what it omitted. Third: restore dignity to slow time —read a chapter together, annotate it, and discuss it for two hours without rushing.
For businesses
Three priorities that go beyond the minimum required by the AI Act. First: include problem framing , or problem formulation, before technical training on the tools. A team that doesn't know how to write a brief won't know how to write a prompt. Second: introduce critical review rituals for AI outputs , with a dedicated human reviewer and traceability of changes. Third: measure the quality of thinking through indirect indicators , such as the quality of internal briefs and the depth of objections raised in meetings.
AI as a cognitive training ground: the paradigm we need to reverse
There's a path that few trainers are seriously exploring: using artificial intelligence not as a shortcut but as a thought coach.
An LLM can play devil's advocate on a strategic decision better than a colleague who fears conflict. They can generate five objections we hadn't considered. They can force us to clarify the implicit assumptions of an argument simply by asking. They can ask us to formulate the problem better before responding.
It's a paradigm shift. Stop asking AI to "give me the solution" and start asking it to "explain to me why my idea might not work ." It doesn't replace thinking. It forces it.
Those training AI agents for Italian companies today shouldn't be teaching prompt engineering. They should be teaching more rigorous thinking, using AI as a constructive adversary. The end product isn't better AI output. It's a sharper professional.
Frequently Asked Questions - FAQ
What does the AI cognitive gap mean?
It's the growing inequality between those who retain attention span, deep reading, and critical reasoning skills and those who are losing them, regardless of their technical proficiency in using AI tools. The gap isn't measured by access to technology, but by the quality of thinking brought to the technology.
Does artificial intelligence really worsen critical thinking?
Not in itself. AI reveals and amplifies a pre-existing cognitive deficit, created by a decade of overexposure to hyperstimulating content. A trained thinker uses AI to think better; an untrained thinker uses it to not think at all.
At what age does the cognitive gap form?
The first measurable signs appear between the ages of eight and thirteen, when average screen exposure exceeds six hours a day and coincides with the development of the prefrontal cortex. But the erosion is also observed in adults between the ages of twenty-five and forty, so it's not just a generational problem.
What does the AI Act say about AI literacy?
Article 4 of Regulation (EU) 2024/1689, effective February 2, 2025, requires providers and users of AI systems—including companies using ChatGPT for editorial purposes—to ensure their staff has a sufficient level of literacy, including the ability to recognize the risks, opportunities, and limitations of the systems used. In Italy, Law 132/2025 implemented the regulation, designating the National Cybersecurity Agency as the competent authority.
How do you train critical thinking in the LLM era?
Three practices converge. First: read long texts with uninterrupted attention, for at least thirty minutes at a time. Second: systematically check the sources cited by the AI before citing them yourself. Third: use LLMs as generators of objections and counterarguments rather than as providers of answers. Ask why my idea might not work instead of asking what the answer is.
The public debate continues to ask whether artificial intelligence will make our children more or less intelligent. That's the wrong question. The right question is: will we be able to develop the mindset needed to use it?
The answer depends on the choices made today, in families, schools, and businesses. It's not a technological issue. It's an educational, organizational, and political issue. The 2030 divide is being written in 2026.
Which side do you want to be on?
Sources
OECD-PISA 2022, Country Notes Italy (December 2023). • Oxford University Press, Word of the Year 2024: brain rot. • Regulation (EU) 2024/1689 (AI Act), art. 4. • European Commission, AI Literacy Q&A (13 May 2025). • Law 132/2025 implementing the AI Act. • AIE Observatory on Reading, presented at Più libri più liberi (December 2025). • National Guidelines for the First Cycle of Education 2025.
Comments