Since the day ChatGPT entered our lives, more and more journalists have, quite curiously, adopted the very same writing “habits” they used to avoid. They suddenly began using semicolons, em dashes, bullet points, and more subheadings. Instead of writing “U.S.”, they started writing “United States.” They stopped making spelling or syntactical mistakes. Their texts became structurally smoother, but also noticeably less personal and less targeted. Without wishing to sow discord, I openly accuse countless colleagues of no longer writing the stories they sign. Machines write them on their behalf. A similar trend can be seen across many professions—people exchange AI-generated texts that they have not even bothered to read themselves.
Emeritus Professor of Linguistics at American University in Washington, Naomi S. Baron, approaches the issue from the opposite angle: how the fiery yet lifeless pens of Artificial Intelligence affect reading and, ultimately, thinking itself. Her latest book, Reader Bot: What Happens When AI Reads and Why It Matters (Stanford University Press), explores what happens to the reading process when humans authorize AI bots to do the reading for them. She kindly spoke with Economic Review about a troubling present and an unsettling future.
Interview by Panos Sakkas
(Published in Economic Review - December 2025)
Creativity, Reading, Writing
How is the growing penetration of artificial intelligence into written language changing the way we create, but also consume, intellectual works of others?
Various AI tools are constantly improving in terms of the kinds of writing they can produce, and I believe creativity is a function of both process and product. Therefore, if we attempt—as many AI researchers do—to evaluate creativity, we can take two texts and ask five or five hundred people to read them and guess which one was written by AI and which by a human, while also asking them which they found more creative.
There was a time when the answer was that human-written texts—even when readers did not know which was which—were considered more creative. Today, there are experiments showing that creativity is often rated higher in works that are revealed to have been produced by artificial intelligence. The question is: what do we mean by “creativity”?
What do we mean, then?
Let us consider a spectrum of writers. On one end, James Joyce. On the other, Ernest Hemingway. Which of the two was more creative? They had different goals. They wrote for different audiences, in different eras. And we, as readers, sought different things in their writing, whether at the time the works were created or decades later. So the way we judge creativity ultimately leads us to the question: what counts as art?
Think about Impressionism and how negatively critics initially received it. They said the works were merely impressions, not real art. Yet today those works are extraordinarily valuable—not only financially, but also as experiences for us as viewers.
Now let us move to creativity in relation to what AI does as a “writer” for us as “thinking readers.” To the extent that we use AI to read for us—say, to summarize technical articles or analyze literary texts—we are making several assumptions. First, that what AI produces is roughly as good as, or perhaps better than, the work we would have done ourselves.
The question is: what do we want from reading? And if you asked me what the most important conclusion is that I have reached and that I try to develop in my new book, I would say that we risk diminishing the importance of reading in many ways. We diminish it when we say that reading for pleasure is not worth the effort and prefer to scroll through Instagram. We diminish it when we forget that reading is both an intellectual and an emotional act. We lose the cognitive benefits of the activity and the potential emotional rewards when we do not read ourselves but instead rely on AI to do it for us.
AI has access to vast amounts of data and may, in some cases, write almost well. But does it possess a sense of purpose in its writing in the way a human creator does? James Joyce, for example, had a story in mind. He wanted to take it somewhere. He wanted to say something. Does AI do that? Does it have that purpose or capability?
The short answer is no.
There are many claims that AI imitates thinking, imitates emotions. Imitation is not the same thing as experience. And it is absolutely clear, to anyone willing to speak honestly, that the sole purpose of AI—at least large language models—is to make good predictions about which word should come next.
The goal is not to feel, nor to write something profound. The goal is to produce text that users find useful and that generates revenue for major technology companies. That is the purpose of the large language models built by Meta, Google, OpenAI, and Anthropic. The objective is profit. Not thought, not creation: profit.
From Human, for Humans? Not anymore
How, then, is our digital life affecting our reading habits and our ability to think, concentrate, and even communicate? What is at risk here?
Let me begin with communication because it concerns me deeply—whether we are talking about predictive texting, Gmail’s Smart Compose and Smart Reply, or now Microsoft Copilot writing emails for us.
I am deeply worried about what happens when we, as human beings, no longer consider it necessary to put ourselves into the messages or emotions we express to others.
There is a study conducted at the Georgia Institute of Technology a few years ago, which I discuss in my new book. Researchers used an algorithm—a precursor to Smart Reply—to generate email responses. The question was: what did recipients think of the sender?
The conclusion was this: if recipients believed the email had been written by AI, they viewed the sender as more rigid, less expressive, and certainly not someone with whom they would want to spend personal time.
Even when researchers falsely told participants that an email had been written by AI when it had actually been written by a human, participants perceived the response as insincere—as something that did not reflect the level of care they expected.
We therefore hold the belief that matters of communication ought to remain between human beings.
As for thinking, a number of experiments published only recently have examined what happens when we ask AI to do something for us—whether writing an essay or using Gemini to gather information for an essay rather than using a traditional search engine.
The conclusion was that when people are required to conduct the research themselves and evaluate the evidence and sources, the final result is clearly superior to what emerges from a Gemini-generated AI overview. The quality of every essay deteriorated the more heavily people relied on AI.
Moreover, participants’ ability to recall information and evaluate it critically was significantly reduced when they had not performed the mental groundwork themselves.
Over the last three or four years, more and more studies have been published showing that we need to be careful what we wish for.
The Generation That Never Had to Struggle
So there is a risk of losing skills because of inadequate research and the absence of human thought? It seems that more and more often, instead of thinking, we simply ask for things to be done automatically for us.
Exactly. And it is not only that.
People who are now in their thirties or forties were once forced—before AI—to develop certain skills. Are they losing some of those abilities now? Probably. It is like a foreign language: if you spoke it fluently until age fifteen but then stopped using it, you are probably no longer very good at it.
My even greater concern, however, is for children in elementary school, middle school, and high school who have relied on AI from the beginning. They do not write on their own, they do not read on their own, and therefore they never develop those skills in the first place.
For these age groups, it is not even accurate to speak of skill loss, because they never had the opportunity to acquire those skills.
Voluntary reading rates had already begun declining worldwide years before modern AI appeared—but artificial intelligence is pushing them over the cliff.
To what extent has this culture of AI convenience penetrated our lives? Will this new AI culture dominate?
Let us take age as an example.
It is easy to speak about digital natives versus digital immigrants—the older generations who use technology but did not grow up with it. Reality is more complicated.
Many people in their forties, fifties, or sixties are far more technologically capable than their children or grandchildren. So it is not simply a matter of age.
That said, it is true that attachment to social media—which also relies heavily on AI, although AI is not limited to social media—is much stronger among younger people, from children to people in their twenties.
A particularly interesting experiment is Australia’s ban on social media for children under sixteen. If you read interviews with teenagers about how they felt before the ban, some said: “This is terrible. They can’t do this to me. My life will be ruined.”
But many others said: “You know what? I have wasted far too much of my life on Instagram, TikTok, and YouTube, and I need to reclaim it. So I’m okay with this.”
We see the same attitude in the United States, especially among university students. Some say: “AI can write for me, but I don’t want it to. It might do a better job and maybe get me a better grade, but that’s not me. I care about writing it myself.”
The issue I have focused on for the last two years is reading, interpretation, and thinking. It is one thing to present a text as your own when it is not. It is another to hand over the function of thinking to AI and lose the opportunity to think.
Modern Loneliness
And in the future?
It is difficult to know what will happen.
Let me use an analogy. In the United States, when Trump was elected for a second term, it took very little time for the culture of the country to change. Who would have imagined that we could lose some of our fundamental values so quickly?
If you asked me to predict today what role AI will play in our lives five years from now, and if I were Sam Altman, I would say AI will be everywhere.
But my answer would be different if I were one of the people who worked for him and chose not to continue, or if I were Geoffrey Hinton—the man who, in many ways, helped set all this in motion and who now leads the experts warning that we should not be conducting this type of AI research because it may ultimately turn against us.
How many people take him seriously? It may take some major accident before we understand. It sounds terrible, but that is often how we learn our lessons.
Does modern technology create conditions of loneliness?
One of the things digital technologies have done is undermine human interaction.
They have also undermined interaction with the outside world. If I go for a walk in the forest and all I see are squirrels and deer, that is not human contact, but I am still interacting with something alive. I am interacting with something that affects me.
I fear that AI, through all the ways it permeates our lives, is turning us into isolated beings rather than social beings living in a real environment.
A Double-Edged Sword for Creators
How do you assess the fact that major organizations are signing agreements with OpenAI, allowing large language models to train on human-created texts, while artists and scientists demand protection of their intellectual property?
I laugh for two reasons.
First, because I recently wrote about this. Second, because Anthropic has just agreed to a $1.3 billion settlement to resolve a major class-action lawsuit.
Recently I received an email from one of the law firms handling the case, informing me that if any of my works appeared on the list, I might be entitled to compensation. They never asked permission—they simply took the material.
How do I feel about that? It is not about the money. It is about the fact that they had no right to take my work.
On the other hand, some publishers I work with have signed agreements with OpenAI—and perhaps other companies as well—making my works available to them.
On one hand, I own the copyrights. On the other hand, if these large language models are used by vast numbers of people to learn things and discover authors, then if they do not have access to my work, people may never discover it.
If I ask ChatGPT, “Who writes interesting books about artificial intelligence and writing?” and my book is not among the sources it draws upon, then someone trying to understand what is happening in this field may never encounter my work.
For an author, that is truly a double-edged sword.
Indeed...
And one reason many publishers have signed these agreements, especially with OpenAI, is to make their authors’ works discoverable.
Take The New Yorker or The Atlantic, for example. These prestigious publications have, I believe, signed agreements with OpenAI because they want their articles to appear in searches—and that makes the publication more valuable overall.
So the decision is not as simple as “do it” or “don’t do it.”
In the lawsuit brought by The New York Times against OpenAI and Microsoft, some examples show GPT producing passages that are almost identical to material from The New York Times, reproducing sections nearly word for word.
OpenAI argues that text is absorbed into databases containing billions or trillions of expressions and that nothing so similar to the original should ever emerge. But that is not always true, and we see the same phenomenon in art.
AI can learn to write in someone’s style. And that is a form of undermining.
The same applies to painting: you look at a painting and wonder whether it is really by Cézanne. The same applies to music: you hear something and ask yourself whether it is Bach—or perhaps not.
Ultimately, how much we choose to rely on technology is something that should be guided by our own human judgment.
(Photo by Ed Feingersh/Michael Ochs Archives/Getty Images)