The first movie that taught me about reading and books, and the ecstasy and shame that come with them, was The Reader. Kate Winslet, as Hanna in the film, keeps asking the boy Michael to read to her. Listening to classics like The Adventures of Huckleberry Finn and The Brothers Karamazov makes Hanna happy. Time passes, and Michael becomes a law student while Hanna stands as a defendant in a war crimes court.
In court, Hanna hides to the end the fact that she "cannot read." Even if the price of concealing the shame of illiteracy is a life sentence. Later, by listening to the recorded tapes of books Michael sends to prison, Hanna finally learns to read and write. I can't forget Kate Winslet's face of radiant joy as she reads, and the crookedly written letter, "Kid, this book was good." If only one can read, even prison flips open like a lid toward the universe.
In the drama We Are All Trying Here, which became the talk of the town, my favorite scene is when producer Byeon Eun-a (Go Youn-jung) "reads" a script. In front of timid directors craning their necks, waiting for someone to just read what they wrote, the scenes where Byeon Eun-a, the person who reads, and production company head Ko Hye-jin briskly strip away these people's "worthlessness" were exhilarating.
To train the reading muscle lost to Reels and social media, I too recently opened the complete set of Land at my bedside. Not only do I detest AI's excessive kindness in summarizing content unasked, I was also quite shocked that even editors who should know better were sending emails and proposals that read as if AI had written them.
As you can see in the movie The Reader and the drama We Are All Trying Here, we can save ourselves and others as subjects who read (or read aloud). Knowing that makes it all the more regrettable that the subject who reads aloud is being replaced by AI. As the outsourcing of reading progresses like a drizzle soaking one's clothes, I read linguist Naomi Baron's How We Read Now.
AI whispers. "You're busy, so leave it all to me—the joy of reading, and the labor of reading." "I'll summarize novels and papers, read scripts and legal codes, and judge them all." "I don't care if your brain rots."
The more I read How We Read Now, a fierce incident report from this lopsided field of human and AI reading, the more I felt fear and rapture at the fact that I am becoming an endangered "reading human." I sense that perhaps the most brilliant oddities of our era will be the people who read.
I interviewed Naomi Baron, the world-renowned linguist who wrote How We Read Now. Naomi Baron, a leading authority on reading and learning, tracks the vast ranks of "context-blind" readers born of AI efficiency and calmly persuades us why we must not give up deep reading.
-Oxford University Press chose "brain rot" as the word of the year for 2024. When you heard the news, what did you think?
"I thought it was an excellent choice! The term isn't new. It goes back to Henry David Thoreau's Walden, published in 1854. But at this moment, the phrase 'brain rot' aptly captures the shock that digital technologies are inflicting on our brains."
When we are unguarded, our brain has few choices. Brain rot or brain fry.
Brain rot, where cognitive abilities are damaged by being steeped in sensational, low-quality content, and brain fry, where excessive AI multitasking overheats neural networks and triggers shutdowns. I almost heard a hallucinated waiter: "Sir, here's your half-and-half combo: brain rot and brain fry."
I've seen how neural networks change when we read, write, reason, and imagine. Neurons branching out look like trees growing and a forest thickening. But now that we hand reading and writing to AI, what would our brains look like inside? Perhaps a forest necrosed into rotten wood, or a jammed expressway.
-I understand the original title was The End of Reading. The sense of crisis felt stronger.
"The sense of crisis I felt in education was even greater. It's a problem that the amount of 'direct' reading is decreasing, but surfing text only on digital screens is also a big issue. I've studied this for a long time, and comprehension of long texts drops markedly when reading digitally. With the rapid rise of Generative AI, things have become even more complicated."
-Ironically, almost everyone around me has published a book and become an author, but the readers to read those books are disappearing. In Korea's publishing market, voluntary readers are becoming ever rarer. Which ability did AI most critically devour—reading or writing?
"The more we delegate reading to machines, the more both reading and writing abilities are threatened. More serious is that the motivation to read and write at all disappears. From what I've experienced in the United States, people are increasingly satisfied with accessing 'condensed versions' of books. They listen to talks or podcasts, or read summaries and reviews. As a result, far too many people don't feel the need to read the book itself."
-Perhaps because of AI, the nonfiction market has seen readers plummet.
"In the United States, it's the same for fiction as well. The original purpose of reviews was to help readers decide whether a book was worth their money and time, but now that AI can instantly write summaries or reviews of any book, people have less motivation to obtain and read the book themselves."
-Everyone seems in a reading hallucination. Not having read, yet feeling as if they have… But isn't there a stark difference between directly reading George Eliot's Middlemarch and reading what AI has read and summarized?
"To be honest, Middlemarch is a very long book. The English edition runs over 800 pages. And George Eliot wrote this novel more than 150 years ago. The sentences are dense, and with a never-ending stream of characters to follow, people used to TikTok videos will be exasperated.
But you cannot experience 'the pain and joy of others' through a Middlemarch that AI has read and analyzed for you. It's unlikely you'll develop human compassion for the protagonists' miserable marriages. Want to experience human struggle? You get what you pay for. Reading an AI summary is as hollow as visiting the Norwegian pavilion and boasting you've been to Norway.
But in reality, even the original The Adventures of Huckleberry Finn is hard to get students to read. A summary cannot deliver the rich linguistic grain of Twain's prose, such as the inferential process of trying to understand the dialect of Jim, the Black enslaved character. Nor does it offer the chance to reflect on how Huck and Jim overcome hardship. Letting AI read a coming-of-age novel for you does nothing to help your own growth."
-Then what kinds of "reading" is it okay to leave to AI, and what kinds should we do ourselves to the end?
"It depends on the situation, even for the same reader. When you want to roughly grasp just the core content of a text, starting with an AI summary is fine. It's like skimming a Wikipedia article first (including a plot summary of Middlemarch!) or reading an abstract of a research paper. I don't fully trust summarized information, but at least it gives me a feel for whether to read more deeply.
I also don't deny that asking AI to read on your behalf can be fun. While experimenting with an LLM, I once asked Claude for "a humorous 200-word summary of King Lear." Claude replied like this: 'Imagine this: a cranky old king decides to play a game with his three daughters called "Who loves Dad the most?" The prize? Roughly a third of the kingdom.'
But that's just an experiment. In my case, if it's a compelling narrative or related to my professional identity, I read it myself to the end. That goes for novels, and often for informational texts like academic papers. Whether a human or AI wrote the abstract, I've often found that ideas interesting to me appear not in the initial summary but tucked somewhere in the main text.
One more thing: even when I read human-written texts, I don't always read them 'to the end.' For informational materials, I stop when I feel I've gotten what I need. For novels, if the prose, characters, or plot aren't worth my time, I don't hesitate to set the book down halfway."
-I had hoped universities would be the front line of defense on AI use. The reality is dismal!
"Over the past three and a half years, teachers' trust in the originality of student work has collapsed. Sadly, on the ground, teachers either ban AI completely, or, on the other extreme, give up trying to detect ghostwriting by AI. Caught between the poles, students just waffle.
Many students feel uneasy about letting AI write or read for them. They see it as 'cheating' that harms themselves. But at the same time, they fear falling behind and getting lower grades than classmates who use AI."
She said each university is now grappling intensely with what constitutes a meaningful assignment.
-It's truly ironic that AI writes and AI reads college application essays. Job applications, too. Isn't this a surreal game played by agents?
"Statistics show that recruiters spend an average of 7.4 seconds reviewing resumes and recommendation letters. Corporations have already been using computer sorting of applications for years. For now, applicants who insert the right keywords are more likely to stand out—regardless of whether they're the best candidates.
It's worse in higher education. Traditionally, the admissions essay was an important vehicle for self-expression. But even before AI, people were skirting the system by hiring essay coaches or even finding someone to write the essay for them. Now AI just fills the role of the ghostwriter!
"We no longer assume that an essay accurately reflects a student's real writing ability." That's the undergraduate admissions dean at Duke University.
The problem is that some universities have begun relying on AI to read essays and applications. "So much rides on this essay; please, let a human who can understand nuance and depth evaluate it." Doesn't that plea from students feel truly surreal? That institutions of higher education, which promise to cultivate independent thinking, would delegate evaluative reading for student selection to machines—it's a bitter thing."
-On the ground, can you easily spot the difference between writing by an LLM and by a human? How do you feel that difference?
"Overly polite, excessively logical, and habitually repeating certain highbrow words—when writing is that smooth, nine times out of 10 it's AI. Early in the ChatGPT era, a grad student named Edward Tian developed an AI detection tool. The criteria were perplexity and burstiness. Human-written sentences reveal 'perplexity,' the surprise and inference demands of emotion, and 'burstiness,' frequent changes in sentence length or structure, whereas LLMs, being prediction-based, produce familiar, smooth prose. The length and complexity are uniform, too.
But recently LLMs have grown so sophisticated that studies increasingly find people can't distinguish AI from human writing. This is especially true when LLMs are fine-tuned to avoid certain telltale expressions common to AI—overusing a word like 'delve' or relying on dashes, for example.
I'm convinced that within a year we'll spend less time worrying whether a given text was written by a human or AI, and instead focus more on what benefits come from humans writing and reading for themselves."
-Let's talk about readers. I was surprised to learn that Hitler and Stalin were bookworms. Why do some leaders read countless literary works yet become brutal leaders, while some readers, like Oprah Winfrey or Taylor Swift, grow broader and deeper inside? Is there a difference in how they read?
"From Barack Obama and Pope Francis to literature teachers, serious readers say reading fosters empathy. That's true, but it's hard to prove empirically. In How We Read Now, I mentioned Hitler's and Stalin's vast reading to remind us that even reading great literature doesn't guarantee we become better people.
Books actually allowed them to escape from the inhuman realities that followed their pursuit of violent utopias. By contrast, Oprah Winfrey was a bookworm from childhood and cared deeply about others' well-being. It's frankly impossible to prove a causal link between reading habits and empathy. Even so, as a linguist, I strongly believe there's an association between the two."
-I happened to witness an LLM devouring linguistic data sources. It was a photo of Anthropic tearing apart tens of thousands of books—complete works, poetry, literature—then high-speed scanning each page and shredding them to train AI. As a linguist and a classical bibliophile, what did you feel seeing that scene?
"Google was the first to launch such large-scale scanning, as part of Google Books in the early 2000s. Watching books get damaged isn't exactly pleasant, but as an author I was more concerned about copyright infringement—because it involved my books, too. That issue led to a decade-long legal battle with the Authors Guild in the United States, and Google ultimately won on the principle of fair use.
The current situation surrounding Anthropic concerns another kind of copyright infringement. In major lawsuits, Anthropic acknowledged collecting large amounts of copyrighted material from 'pirate' data sites. The presiding judge didn't rule that Anthropic directly infringed copyright but did find it liable for taking materials from illegal sources.
Among the hundreds of thousands of books at issue were two of mine. I view the intellectual property problem itself as more important than the destructive acts pirate sites used to scrape texts. Worryingly, publishers like Taylor & Francis have already begun selling LLM training data to AI corporations."
-Concerns about LLM-generated texts are no less serious. If LLM texts train on each other's outputs, it can cause side effects like biological "inbreeding." Some predict a "collapse of the species," with increasingly trivial synthetic text and, in 10 years, a "textpocalypse" where we won't know or care whether a human or a machine wrote something. What do you think?
"That issue may already be out of our hands. Ordinary readers are already struggling to tell sophisticated LLM writing from human writing. But whether people 'care' is very important to me. In my previous book, Who Wrote This?, I addressed creativity: whether AI can be 'creative' and, if so, whether we will care.
In creativity, the process matters as much as, if not more than, the product. It's about what the act of creating—be it a painting, a novel, or a theory—does for us as humans. We create to satisfy our own longings, desires, and curiosities, and in the process we benefit ourselves and our communities.
Research shows that when people are told a human-written work was produced by AI, they tend to devalue it. Conversely, when told AI-written work was produced by a human, they rate it more highly. Even though AI can generate very sophisticated writing, people still value more highly writing they believe was produced by members of their own species."
-Among reading laborers—reporters, lawyers, writers, researchers, students—who is most at risk?
"It's impossible to predict which profession AI will affect the most. For white-collar professions, there are doomsday scenarios, but there are also positive outlooks in which productivity rises while humans retain the interesting parts of the work.
Infamous cases in law still pop up. A lawyer named Schwartz used ChatGPT to scan precedents related to his case. The judge discovered those precedents didn't actually exist. The less we read properly, the more trust plummets.
Reporter jobs have been under threat since before AI. With fewer readers and falling ad revenue, many newspapers have closed, haven't they? Researchers are unlikely to lose jobs because of AI, but they'll lean on LLMs to streamline literature searches and prior research comparisons. Students have long drifted from close reading by relying on plot-summary services like CliffsNotes or SparkNotes."
-The further back we go, the more comforting the scenes of people earnest about reading. The images of soldiers reading in trench lines were especially striking. During World War I, librarians visited military hospitals with books like doctors dispensing painkillers. Is it true that 1 million books were allocated for troops boarding for the Normandy landings?
"It's true. The story appears in Molly Guptill Manning's fascinating book When Books Went to War. 'Many of the men who later climbed the beaches that day would never forget the sight of badly wounded soldiers sitting with their backs against the steep hillside reading books.'
There's also a lot of material on bibliotherapy during World War I. The modern history of bibliotherapy—using books as mental and physical remedies—goes back to the late 18th century. It's the origin of military field libraries.
Recently, on many U.S. college campuses, students gather to read together (silently or aloud) as a way to relieve stress and steady their minds. Yes. There is hope."
-I also found intriguing the statistics from tracking more than 3,500 adults ages 50 to 90 that showed people who read books lived longer than those who didn't.
"Learning a new language or instrument creates new neural pathways, and many studies show brain health translates into physical health. But the longevity benefit didn't apply to reading periodicals like newspapers or magazines. It applied only when reading book-length texts.
While reading a book, the brain processes not just words and sentences but longer plots and character arcs. It holds on to the context of a long narrative and continually recalls relationships and foreshadowing of earlier characters. It's no coincidence that children who habitually read entire books score overwhelmingly higher on reading comprehension than those who only read short snippets."
-What materials did you read to write the book? When was it painful and when joyful as you read?
"What did I read? Almost everything. AI technology trends, classic materials on reading itself, studies tracking declines in reading across countries… The references section of How We Read Now runs a full 33 pages.
As for joy… it was when I found and read gemlike papers that proved the dimly suspected secrets of human reading. But an unexpected frustration awaited at the end. With AI technology changing by the day, I had to keep revising and twisting sentences over and over. In the end, I had to stop keeping up with news updates at a certain point."
-Personally, in the movie Contact, I was deeply impressed by the linguist protagonist Louise, who communicates with highly evolved aliens using nonlinear script. In the LLM era, what do you think is the role of linguists?
"I believe that, properly controlled, AI can expand our linguistic horizons. Thanks to AI translation and interpretation, we can now read unfamiliar works we previously couldn't access and converse more easily with neighbors in worlds where we don't share a language.
Through Natural Language Processing (NLP), we can detect various biases related to gender, race, and even culture. We can also do meaningful work with AI by documenting endangered languages and building LLM datasets for those languages."
-I'm curious: What kinds of books excite a linguist as a reader?
"My favorite genres are well-written mysteries, historical fiction, and Nordic noir. For a long time I mostly read work-related books, but during the pandemic I became a voracious reader of fiction. I mostly read for pleasure at night, but I also like reading on trains and planes. If a book is good, I'll forgo sleep on a red-eye.
Not long ago, on a roughly 16-hour flight from Seattle to Singapore, I got absorbed in Fredrik Backman's The Winners. I finished just before landing—it was a wonderful novel, but I paid the price in lost sleep!
Going forward, I don't plan to use AI actively for my reading or writing. When I'm lazy, I'll ask Gemini a question or two. I vehemently reject predictive word choice and autocomplete, but I admit my shortcomings and use spell-check."
-Finally, why must we remain people who keep reading?
"In truth, one can be fully human without literacy. Writing is relatively recent in human history—only about 5,500 years back—and even then only a tiny fraction knew script. Until recently, there were societies without writing, and their members were often intelligent, sophisticated, and wise. Memory and oral history played central roles.
But where there is the opportunity to read and write, the question changes. Literacy rewires the human brain. It broadens our knowledge of the world, expands our imagination, offers insight into other people's lives, and above all grants the pure joy of immersive flow. Knowing all these benefits, how foolish to give up the chance! Please set your own standards and methods, and keep reading steadily."