Tom Zeller Jr. is the editor in chief of the digital science magazine Undark and author of “The Headache.”
In her 2025 book, “Humans: A Monstrous History‚” historian Surekha Davies recalls arguing with a graduate student over the virtues of artificial intelligence-generated poetry. The technology, she’s told, is a wonderful advance because it relieves humans of the toil required to sculpt engaging verse. Now, the student said, people won’t have to do it.
Davies recoiled. It was one thing to build machines to clean up nuclear messes or perform other mindless tasks. But to get robots to write poetry ‘so that we don’t have to,” she lamented, “seemed a toe dip in a new pool of dangerous waters.”
There is a lot of toe-dipping going on these days. Last month, journalists discovered that “The Future of Truth,” a book about AI’s effect on knowledge, contained manufactured quotes and other inaccuracies that the author had apparently copied and pasted from a large language model. At around the same time, it appeared as if one of the prizewinning stories published in Granta, a prestigious British magazine, was written by a bot. This spring, Hachette announced it was canceling Mia Ballard’s novel “Shy Girl” in the U.S. over similar allegations.
All this — uncertain authorship, neutered prose, the disintegration of trust about who’s written what — has profound consequences. But the scandals raise a more unsettling question: What happens when we begin to outsource one of the brain’s most cognitively integrative activities?
It’s true, for all but two or three centuries, few humans could or even needed to write. In recent decades, however, neuroscientists have begun to discover that even simple writing tasks can activate a unique and nourishing complex of cerebral anatomy. Such activity unites brain centers that govern language, memory, motor coordination and perception. Composing sustained text, it turns out, isn’t merely a transcription exercise converting our inner thoughts into shareable glyphs. It is a whole-brain workout.
Handwriting analyses are especially instructive. In one brain-mapping study out of Norway in 2024, researchers demonstrated the rich and sustained linking of disparate cerebral anatomy in volunteers tasked with handwriting exercises. The electrical rhythm of their brains showed a deeper, more cohesive synchronization than in those asked to type the same words. This sort of “brain connectivity,” they argued, is the hallmark of a neural state primed for learning and encoding new information.
Researchers in Beijing noted a similar effect in 2022. Children between ages 9 and 11, the authors found, had already recruited an adultlike network of motor, parietal, visual and cerebellar regions when handwriting Chinese characters. Kids whose brains showed stronger handwriting‑related activity in these movement and visual processing areas also tended to recognize and read characters more quickly than their peers.
These findings underscore what many have already discovered in the classroom: Generative writing — putting pen to paper in your own words — tends to produce better recall and deeper learning than, say, verbatim note-taking. A growing body of research also suggests that expressive composition is associated with a variety of psychological and even modest physical benefits.
Then along came AI.
ChatGPT debuted in November 2022, and four years later, the majority of high school students report using OpenAI’s model to help with assignments and homework. One study published in April estimated that as much as 35 percent of new web content is AI-generated, and many white-collar workers report using the technology at least sometimes to draft and revise text. Even more disturbing, at least a quarter of singles reportedly use it to help with their dating lives, from polishing profiles to crafting solicitous messages to potential paramours.
Such rapid and pervasive adoption suggests that AI is something humanity very much wants. But what makes this transition unusual is that writing isn’t merely a productive output. We aren’t replacing hand-stitching with sewing machines or swapping the washboard and tub for a Whirlpool appliance. Composition provides a unique form of cognitive exercise, which some fear AI might be automating away.
The early evidence is worrying. Researchers at the Massachusetts Institute of Technology recently demonstrated that the chatter between interconnected brain regions appears to grow quieter with each additional layer of technological assistance. Asked to write a short essay in response to a prompt, unaided by search engines or AI, the brains of volunteers lit up as electrical and chemical information zipped along intricate webs of the cortical superhighway. Another group, asked to perform similar tasks with the help of an ordinary search engine, still saw plenty of brain interconnectivity but measurably less than the first group. A third cohort, permitted to use AI, showed the weakest connectivity and had the most difficulty recalling their own essays.
One overlooked finding: Participants who had previously relied on the LLM and were then asked to write responses to similar essay prompts without AI assistance still showed weaker neural connectivity than their counterparts who had never used AI. That might sound unsurprising. GPS, calculators, spellcheck, and myriad other tools have sometimes been shown to have detectable, if modest, effects on spelling, math and spatial skills. But the MIT results suggest something more stark: The diminished brain interconnectivity associated with LLM writing assistance might persist even after the crutch is taken away.
The MIT team acknowledges that more research is necessary, but if its findings hold, the worry is not simply that adult users will become cognitively lazy. It’s that many younger users will struggle to develop the sort of rich, interconnected neural kingdoms that writing tends to build.
I find LLMs an incredibly useful tool for interrogating ideas, compiling evidence, testing assumptions, identifying holes in my logic and even proofreading my work. I don’t like the way they edit me, even when the suggestions are obvious improvements, and there’s the rub. AI can write, often nicely, so there’s little mystery behind its appeal. Writing is hard, and many people dislike doing it. Evidence suggests that even before ChatGPT, young people were already writing less frequently for pleasure.
Having an eloquent wordsmith in our corner might prove too alluring to resist, and perhaps that isn’t so bad. With the right safeguards, and with rules dictating that we acquire at least some compositional proficiency before offloading the hard work to algorithms, maybe AI will come to represent more of a “negotiated collaboration,” as some researchers have imagined. Maybe it will even unlock new wells of human creativity and different but still-nourishing cognitive networks.
That would be welcome. But there’s also the possibility that in our rush to automate the burdens of writing, we lose a valuable form of cognitive strain. And a culture that no longer wrestles its thoughts into structured prose, however clumsily, might find that it has fewer thoughts worth wrestling with at all.
The post AI is turning to your brain into mush appeared first on Washington Post.
