I’ve been writing about the effect AI is going to have – and is already having – on the world pretty regularly since the beginning of this year. I don’t get a lot of pushback from readers of the essays and articles I publish. But when I make the same comments to business partners and colleagues, I receive a good deal of it.
I can think of a probable explanation. Readers are generally laypeople who are less likely to know very much about the subject and are, therefore, more curious than worried about it. My peers in the information publishing industry are – whether they like it or not – already mired in the discussion because of the threat it poses to their and our financial futures. So they’ve naturally spent a fair amount of their worry time mining for and discovering facts and theories that suggest my predictions are frantic and delusional, or overblown at best.
Not surprisingly, those who disagree with me most strongly are executives, consultants, and “creatives” (as they call themselves these days) who are in the top decile of income earners due to their accumulated knowledge, skillful decision making, and creative thinking. In other words, the people who have the most to lose as AI steadily increases the supply of what they are paid to do, thus steadily decreasing its financial value.
They don’t say so directly, but they imply that I’m acting the alarmist, distressing myself and others about doom-and-gloom scenarios that won’t come to pass. They remind me that all previous technological revolutions have created more jobs than they have displaced and have resulted in a higher GDP for the country and higher per capita income for all.
True enough. But the point I’ve been making every time I talk (or write) about AI is that what it is replacing is not muscle power. It’s replacing human knowledge and thinking with mechanically stored memory and artificial thinking that is currently equal to – soon to be better than – the kind stored in and coming from the human brain.
They disagree. They insist that AI cannot do what they do because AI is not human and therefore does not think like humans do – i.e., creatively and intuitively. And when that doesn’t stop me, they make an argument similar to the one Freddie deBoer made in a recent essay that I critiqued in the Oct. 26 issue – that the things I’m predicting haven’t happened yet.
“Wait and see” is all I have left to say, and so I often don’t say anything.
But just this morning, I saw another account of my same argument in The Wall Street Journal. The piece, “AI Has Plunged the Book Publishing Industry into Utter Chaos” by Anna Silman, makes my case better than I could have, though I doubt that was her intention.
Have You Heard About Daggermouth?
In July, a literary agency backed out of a multimillion-dollar deal it had just landed for a debut crime novel, telling the publishers it could no longer verify that its client had written the book himself. And in March, Hachette canceled the US release of a horror novel over the same suspicion. But Simon & Schuster is standing behind Daggermouth, a self-published dystopian romance it bought for seven figures after the book took off with readers, even though a Stony Brook computer scientist says it bears a machine’s fingerprints.
The authors all deny using AI, and the publishers seem unsure what to believe. One agent told Silman that she and her colleagues are now being asked to act as police.
Silman’s story is one of many. Over the past few months, I’ve seen some version of it in newspapers, literary magazines, Substack newsletters, and the online forums where writers gather to complain about their agents. The names change, but the plot doesn’t. A book is bought, praised, sold, and loved – and then somebody runs it through a detection program and the room goes quiet.
What caught my attention wasn’t the scandal. It was a remark from that Stony Brook researcher, Tuhin Chakrabarty, about why Simon & Schuster wouldn’t drop Daggermouth. His answer, more or less: It was because the book was already a big success. People were liking it.
The 1% Problem
For years, the argument against AI-written fiction was that it was bad. Flat, generic, emotionally dead – “slop,” to use the current term of contempt. And a great deal of it is. Andy Hunter, who runs Bookshop.org, told Silman he figures 99% of the AI books for sale online are junk designed to trick buyers. I believe him. I’ve read some of it.
But the books causing all this trouble aren’t the 99%. They’re the ones that got past the gatekeepers. They were good enough to win over agents whose job is mostly to say no, editors whose careers depend on picking winners, and readers who had no idea who – or what – wrote them.
Daggermouth spent months on the USA Today bestseller list. You don’t get there by tricking people. You get there because the people who read it told their friends to read it, too.
Now, if that were the whole story, my colleagues could fairly say this is genre fiction – romance and fantasy, the kind of books readers devour three a week. Nobody confuses it with literature.
So let’s talk about literary fiction.
“The Serpent in the Grove”
Every year, the Commonwealth Foundation gives a prize for the best unpublished short story from anywhere in the old British Commonwealth. It’s about as literary as literary gets.
This year’s competition drew 7,806 entries. A panel of judges, chaired by the novelist Louise Doughty, narrowed them to five regional winners, which were published in May by Granta, the famous British literary magazine.
Within days, readers began pointing out that one of them – “The Serpent in the Grove,” by a Caribbean writer named Jamir Nazir – read an awful lot like it came from a chatbot.
Ethan Mollick, a Wharton professor who studies AI, ran it through Pangram, the same detection tool used on Daggermouth. It came back 100% machine-written, a result Wired said it confirmed independently. Then people checked the other winners. Three of the five triggered strong flags.
Granta’s response was a little comic. Its publisher said her team had shown the story to Claude, the AI model, and asked whether it was AI-generated. (I won’t make too much fun of that. I’m writing this essay with Claude’s help.)
What happened next is the part that matters. On July 1, the judges gave “The Serpent in the Grove” the overall prize. The chair praised it as original, poetic, and deeply moving.
Think about that if you are an AI Denier!
A panel of serious, experienced literary judges – precisely the people who are supposed to be able to tell the difference – read stories from all over the world and picked, as the best of them, one that a detection program says a machine wrote. And when the accusation came, they stood by their choice.
Maybe Pangram is wrong. Maybe Nazir wrote every word. I’ll come back to that. But either way, the judges couldn’t tell.
This isn’t the first time. In January 2024, Rie Qudan won the Akutagawa Prize, Japan’s most prestigious award for new writers, for her novel Sympathy Tower Tokyo. The committee called it practically flawless. At the press conference afterward, Qudan mentioned that about 5% of the book had come from ChatGPT. Nobody on the committee had noticed. And keep in mind, that was ChatGPT in 2023 – which, by today’s standards, was a bright child with a library card.
Then, this past May, Olga Tokarczuk – the Polish novelist who won the Nobel Prize in Literature in 2018 – told an audience in Poznań that she’d bought the most advanced AI model she could get and was often shocked by how much it broadened and deepened her creative thinking. For literary fiction, she said, it was “an advantage of unbelievable proportion.”
The backlash came fast. Within days, she issued a statement saying she’d used AI only for preliminary research and that none of her texts had been written with its help.
I’m happy to take her at her word on that. But I find her first statement more interesting because it was spontaneous.
What Readers Can’t See
* In a study reported this August, Villanova researchers had 1,682 readers rate short stories – three written by human authors and three AI versions of the same stories. On average, readers rated the AI versions higher for quality and immersion. Telling readers a story was human-written raised its scores, even if a human didn’t actually write it.
* In two related experiments with more than 900 people, readers identified AI fiction at about chance rates (roughly 40% and 52% accuracy). People who described themselves as frequent readers of literature did no better.
* Chakrabarty’s Stony Brook study of more than 14,000 randomly chosen Kindle ebooks found that one in five contained more than 25% AI-written or AI-assisted text. Those books accounted for about 12% of sales and 10% of bestseller positions.
* The novelist Vauhini Vara tested whether people who knew her well could tell her prose from AI imitations trained on her earlier books. They couldn’t.
The Case Against – and Why It Doesn’t Hold
In fairness to the other side – and the other side has some very smart people on it – let me give their best argument. Marlon James, the Booker Prize-winning novelist who also teaches creative writing, put it as well as I’ve heard it put. “AI,” he said, “will never develop consciousness. It isn’t intuitive. It’s doing mathematics. It isn’t feeling emotions.”
I don’t argue with what James says about what AI is. But I think he’s dead wrong about what it can do. He (and his fellow AI Deniers) need to recognize the simplest fact: The fiction reader never meets the author’s consciousness. He meets sentences on a page. If those sentences move him, he has been moved, whatever produced them. You can say he’s been fooled. But then you have to explain why the Commonwealth judges were fooled, and the Akutagawa committee, and those 1,682 readers at Villanova – all people who, by James’s reasoning, should have felt the absence of a mind behind the words.
Their second argument is that the detection tools aren’t perfect. True. Simon & Schuster says they’ve been shown to produce false positives. Some of the writers being accused of using AI to write their books are surely innocent. And being accused of something you didn’t do – publicly, with your career on the line – is a terrible thing.
But my argument doesn’t depend on any one of them being guilty. Whether or not some of the accused really did use AI, the best-trained eyes in publishing can no longer reliably tell a human writer from a machine, and the industry is tearing itself apart trying. I’ve turned this over every way I can, and I can’t find a version that helps my colleagues.
Nobody Fakes a Watch Nobody Wants to Buy
Which brings me to what I find the most telling evidence of all. It comes from the people who are most certain that AI can’t write fiction.
Last year, 70 authors, including Margaret Atwood and Jonathan Franzen, signed an open letter asking publishers to promise they would never release books created by machines. The Authors Guild, the largest professional writers’ organization in the US, has introduced a “human-authored” certification system. And the Commonwealth Foundation is rewriting its rules. Agents are running manuscripts through detection software, and some are refusing to represent anyone who has touched AI, even for research. In China, where readers devour serialized online novels by the millions, the big platforms have started restricting AI submissions. One of them says it rejected more than 104,000 low-quality submissions, AI-generated ones among them, in June alone.
I sold watches once. One thing I learned is that nobody counterfeits a watch that nobody wants to buy. Pledges, certifications, oaths, and detectors are what an industry builds when it’s worried about competition – and you don’t worry about competition from something that can’t compete. Nobody ever asked publishers to pledge that they’d never print a book written on a typewriter.
The most honest line in Silman’s piece comes from an agent who plainly hates AI. The only silver lining, she said, is that publishers are still canceling these books. But she added that she doesn’t know if that will continue. She’s afraid of the day readers stop caring who wrote the book. And judging by the Villanova study, a lot of them already have.
What This Means for the Rest of Us
Now back to my colleagues, the executives and consultants and creatives who tell me AI can’t do what they do because it isn’t human and, therefore, can’t think creatively or intuitively.
Fiction is the hardest test of that claim that I can think of. A good story doesn’t just inform or analyze. It has to make the reader care about people who don’t exist. That’s about as close to pure creativity and intuition as writing gets.
If a machine can produce a story that serious judges call original, poetic, and deeply moving – one that ordinary readers prefer when they don’t know where it came from – I think we have to accept that.
So what should you do if you earn your living with words, as I have for most of my life?
I’m not going to tell you to panic. Panic is not a plan. I’ll tell you what I’m doing instead.
I use AI tools every day, because the cost of not doing so is falling behind the people who do. And I’m putting more of my time into the things a reader pays for that a well-made sentence alone can’t give him: a name he has learned to trust, a point of view he has come to rely on, a relationship built over years of keeping promises. A machine can write a story that wins a prize. It can’t give you a reputation that took 40 years to earn. Reputation compounds, and right now it may be the most valuable asset a writer owns.
I used to tell my colleagues “Wait and see.” I’m going to try something different now. I’m going to tell them to read “The Serpent in the Grove”… and then tell me who wrote it.
Politics
Three Quick Bites
Mary Peltola
Skulduggery in the Alaska Senate Race
On Oct. 21, The New York Times published a story demonstrating that operatives for Mary Peltola, the Democratic Senate nominee in Alaska, recruited a candidate with the same name as the incumbent, Dan Sullivan, to confuse voters and boost her chances of victory. Read this editorial from The Free Press on what has become a familiar pattern: Democrats casting themselves as saviors of democracy, while subverting it themselves.
Elon Musk Is Under Attack… Again
Democrats are admitting they are going to try to put Elon Musk in jail if they take back power in Congress. They can’t tell you what he specifically did wrong, but they’ll come up with something. Just as bad, or even worse, is that some of his critics want to nationalize his companies, including Starlink, which provides crucial infrastructure to Ukraine. This idea is crazy and reckless – and even dangerous. Here’s why.
A supporter of Lula da Silva celebrates after the results of the presidential runoff election on Oct. 25
Lula Won the Presidency in Brazil and Tried to Put His Opponent in Jail
PB, one of my trainers, is delighted that Lula won the election in his native Brazil. He was even happier at the prospect of Lula putting his political opponents in jail. I like PB too much to argue with him. But most of the other Brazilians I know see things very differently. They think Lula is corrupt and fear what another term will do to Brazil.
In this article, Michael Shellenberger goes considerably further. He argues that Lula and Supreme Court Justice Alexandre de Moraes are using censorship and the courts to suppress political opposition. It’s a disturbing case, and worth reading.
Worth Considering
Postscript: What can a human being accomplish in 7 years?
It can do this…
