You Have 18 Months: AI and the Quiet Decline of Deep Thinking

Strength coaches talk about "time under tension." Take a simple weighted squat. One person can lower and raise the weight in two seconds; another can take ten seconds with the same load. The slow version is harder, and it builds more muscle. More time under load means more strain, and more strain means more growth.

Thinking works in a similar way. The ability to sit patiently with ideas that are only loosely connected, or not connected at all, is what lets a person eventually weave them into something new. That kind of slow, effortful combination is hard to observe from the outside, but it is the engine of original thought, and it depends on staying with difficulty long enough for something to form.

That idea is a useful lens on one of the louder claims in the AI debate. Several prominent AI executives and commentators have suggested that humans have roughly 18 months left before AI systems overtake them in the workforce. By around the summer of 2027, the story goes, an explosion of machine capability will leave people behind. Up to half of all entry-level white-collar jobs could disappear, and even the most brilliant researchers might be outclassed by what one AI leader has called a country of geniuses in a data center.

Taken literally, the prediction is wildly ambitious and almost certainly wrong. Yet it resonates. One of the questions parents ask most often right now is some version of this: if AI is about to be better than us at everything, what should our children study? If generative models can outperform software developers at coding, radiologists at diagnosis and mathematicians at problem solving, then even the traditionally "safe" fields such as computer science, medicine and mathematics no longer look safe.

The anxiety is understandable. But forecasting how AI might one day make human workers obsolete is guesswork. Describing what the technology is already doing to our capacity for deep thought is not. The more immediate risk over the next 18 months is not that AI makes everyone unemployed, or that students lose to machines in some head-to-head contest. It is that, surrounded by powerful new tools, people quietly degrade their own abilities. The focus on how technology might beat us has obscured the ways we may be beating ourselves. The decline of thinking people deserves at least as much concern as the rise of thinking machines.

Reading, writing and the thinking they train

Earlier this year, a New York Magazine cover story spread quickly, not because it shocked anyone, but because it said out loud what many already suspected: students everywhere are using AI to cheat. Large language models can produce an essay on any topic in seconds, which has created an existential crisis for teachers trying to assess whether students can actually write. Students described college as an exercise in using ChatGPT well, and professors warned that many graduates will enter the workforce effectively unable to write.

AI Skills for the Modern Workplace: What Knowledge Workers Need

Companies are adopting AI quickly, but many employees still lack the skills to use it well. Three core abilities close that gap: problem framing, critical evaluation and effective prompting.

Vite’s Future Is Paid: A Deep Dive into Vite+

As a Rust-based unified toolchain, Vite+ promises performance and consistency, but its commercial license sparks funding and adoption debates

This matters because writing is not a separate step that comes after thinking. Writing is how thinking happens. That holds for professionals as much as for students. An editorial in Nature, titled "Writing is thinking," argued that outsourcing writing entirely to language models deprives scientists of the very process through which they work out what they have discovered and why it matters. Anyone who hands their writing to AI risks filling screens with words while emptying their own minds of ideas.

Reading is in even worse shape. A university professor writing under a pseudonym argued this spring that many students are now functionally illiterate, and meant it literally. The data supports the concern. After decades of gains, literacy and numeracy scores are falling across much of the Western world, a trend that led John Burn-Murdoch of the Financial Times to ask whether humanity may have passed "peak brainpower" at the very moment it is building machines to think on its behalf. In the US, the National Assessment of Educational Progress (NAEP) reported that average reading scores in 2024 fell to their lowest level in 32 years, which is as far back as that data series goes.

People still read constantly, of course: emails, messages, social feeds, subtitles. But those fragments rarely demand the sustained attention required to follow the argument of a long text. The share of Americans who say they read books for pleasure has fallen by nearly 40 percent since the early 2000s. Even top students read little beyond a paragraph at a time. Reporting in The Atlantic described students arriving at elite universities without ever having read a full book for school, and faculty who find that their students struggle to hold focus through a single sonnet. Education researcher Nat Malkus of the American Enterprise Institute has suggested one cause: schools broke books into short excerpts to prepare students for standardized reading-comprehension tests. In optimizing for the measurement of reading, the system may have accidentally killed the habit of reading books.

Computer scientist and author Cal Newport, best known for Deep Work, describes reading and writing as the twin pillars of deep thinking. A modern economy rewards symbolic reasoning and systems thinking, and nothing trains those abilities better than sustained reading and careful writing. In Newport's view, AI is the latest heavyweight contender against our ability to think. Television coincided with falling newspaper subscriptions and a slow decline in reading for pleasure; the internet, social media, smartphones and streaming each pushed further in the same direction. He compares reading and writing to a superhero serum that grants the power of deep symbolic thought, and argues that we need to keep drinking it.

The warning echoes an older argument from Walter Ong's book Orality and Literacy. For Ong, literacy is not a passing skill but a technology that restructures human thought, making room for ideas too complex to hold in memory alone. Stories can be memorized and passed on by people who cannot read. A work like Newton's Principia cannot be transmitted across generations without written formulas. Oral languages typically work with a few thousand words, while written standard English has accumulated well over a million. If reading and writing rewired the brain's capacity for logic, their decline risks unwiring that capacity just as far larger machines arrive.

So what should young people study in an age of thinking machines? No one can say with confidence which field is safest. The more useful answer concerns which abilities to value, and they are the same ones now in retreat: reading long and complex texts with patience, holding conflicting ideas in mind and enjoying the tension between them, and working closely with language at the level of the sentence. In a world where video replaces reading and generated essays replace writing, choosing these habits deliberately is a competitive advantage. As AI becomes abundant, deep human thought risks becoming scarce, and that scarcity is a clear and near threat.