AI crossed a line and it's nothing like book burning, it's worse.

People who use AI say it's the future. Here's why they're wrong

Book burning photo from Wikipedia

Years ago, I read The Book Thief, by Markus Zusak. Five hundred weeks on the New York Times bestseller list. It sold almost twenty million copies and USA Today said it belongs on the same shelf as Diary of a Young Girl, by Anne Frank.

It opens with Death, saying he likes colors. They’re a distraction that keeps him sane, given how long he’s been doing his job without a vacation and you might wonder what Death needs a vacation from, but he tells you.

The leftover humans.

The survivors.

He can’t stand to look at them. Left crumbling among jigsaw pieces of realization, despair and surprise. Standing there with punctured hearts and beaten lungs.

This is just a small story, he says. About a girl, some words, an accordionist, some fanatical Germans, a Jewish fist fighter — and quite a lot of thievery. The first time he saw the book thief, she stole a book out of the snow. The second time, out of a fire.

I’m seeing that in my feed. Fire. Book burnings. People comparing AI companies to Nazi book burnings. Maybe you’re seeing it, too. And I get the comparison. I get why people are saying that—because of the destruction. But it misses the point. Entirely.

---

Here’s the story that’s all over the news. AI companies are buying used, old and rare books to scan them into AI and then destroy them.

Here’s something you should know. There are two ways to scan a book. One is non destructively. There are machines that scan without destroying. The second way is destructive scanning. Machines slice the spine and binding off a book and run the pages through a scanner. After the pages are scanned, they’re shredded.

Destructive scanning is the cheapest way to scan books.

People are reading that, saying omg, it’s like Nazis burning books. It’s not. People are asking why they are destroying the books they scan but that goes back to the Anthropic trial last year.

Anthropic got taken to court by writers whose books were pirated and fed to AI, so they sued in a class action suit. Like every company accused of breaking the law, they put a suitcase full of paperwork in front of the judge. He read it and lambasted them.

Look, he said, if you want to buy books and destroy them, that’s up to you. That’s not breaking any laws. But you pirated millions of books off the internet and that not okay. That’s piracy and it’s against the law and you’re going to have to pay.

The news stories went straight to the billions in piracy settlements.

No one asked why the judge said if they want to buy books and destroy them, that’s up to them. Too focused on the billions and piracy. Very few people wrote about the mention of Project Panama in the documents Anthropic submitted to the court.

In the court documents, Anthropic talked about a secret project, submitted to let the court know they’re going to start paying for books. Here’s what it said:

“Project Panama is our effort to destructively scan all the books in the world,”
“We don’t want it to be known that we are working on this.”

That might make your heart go cold. It did mine. But it’s not new. That’s was published in The Washington Post last January, six months ago.

That’s what the judge was talking about when he said if you want to buy books and destroy them, that’s up to you. That was last year.

People destroy books all the time. Buy them, read them, throw them in the garbage or toss them in the recycling bin. It’s not illegal. If you pay for a book, you compensated the publisher and the author—or the estate if the author is dead. After that, you can do what you want with it. It’s yours.

No one had any clue what was coming. Because piracy. Because billions!

---

Know what we didn’t see coming? The middle men who crawled out of the woodwork, offering to buy up millions of books so no one will know who bought them.

We aren’t talking back alley shysters. I’m sure there’s some of them, too. But also? Big companies, who saw an opportunity to pad the coffers with AI investment dollars.

For example? ISBNdb is the world’s largest book database. For years, ISBNdb was a godsend to the publishing industry. From the beginning of publishing to the time AI was launched, a total of 130 million books were published. 110 million are listed in the ISBNdb database. 85% of all books ever published since the dawn of publishing are in their database. If you wanted to find a rare book or a book that’s out of print, they were where you would find it. In the biggest book database in the world.

They’re marketing a new service.

Offering book acquisition services to AI developers.

They say printed works published before AI existed are “especially valuable” because there’s no chance they contain AI generated text. AI companies are very specific about that. They don’t want any materials that have any chance of containing AI generated text. You know why? Because AI generated text corrupts the model.

Someone is going to argue with me. Say nuh-uh, they have their ways of training on AI generated text. Tell that to the book dealers spending million on used books for AI companies who don’t want people to know who is buying them.

LLMs are just probability generation, like your phone, except with more words in its belly, but no matter how many words they feed it, it bases the output on probability. So they don’t want to train AI on AI generated text. It degrades the model.

They try to play that down. They don’t really want the world to know text created by their own wonder tool degrades the tool itself. In the beginning of AI training, they could use sites like Wikipedia and news media but they can’t anymore because the risk of corruption by their own text is too high. Yay undisclosed AI, right?

Sometimes, I want to shake my head that this so called “tool” ate so many beautifully crafted words, ate The Book Thief, Hemingway, Kafka, Dostoevsky and Atwood, and it still can’t write well enough to save its own life. It needs more human writing, more, more, because its own writing ruins the tool and only human writing can save it.

While people who love it scream about luddites and tell me this is the future.

Nah. Let me tell you what the future is.

The future is a man standing in front of a judge saying we want to destructively scan all the human authored books in the world, but we don’t want the public to know.

That’s “the future” of AI.

ISBNdb says the orders they’re getting range from a thousand books to a million.

---

The reports are rolling in now, from around the world.

Rare book dealers and secondhand booksellers in European countries and the Netherlands are seeing a wild increase in orders. They’re concerned that AI companies are buying up obscure and rare titles and destroying them after feeding them to AI.

But they don’t know who the buyers really are. So many anonymous companies.

The orders vary wildly in size, which makes it harder to tell. If someone buys a million ISBNs, that’s probably an AI company. But what about the orders for 500 or 100 rare books. Is that a rare book dealer or an AI company? And how are they to know?

What are booksellers supposed to do? Stop selling rare and limited edition books? Hide them to save them from destruction, like we did from the Nazis?

When I read that rare book sellers and used bookstores are seeing a five and tenfold increase in orders due to AI companies buying books for destructive scanning, I can’t help but think of what Sam Altman said at the Blackrock Summit in Washington, DC.

“We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter and use it for whatever they want to use it for.”

Let me tell you something blunt.

When the Nazis burned books, they didn’t hide behind anyone. They didn’t have middlemen skulking around to find the books. They threw books in the street and said we don’t care if you like these books, we don’t so we’re burning them. People fought to save them. Pulling books from burning piles. Librarians hiding books on the list. People risking their lives to save rare books from rampant destruction.

It kind of sucks to say the Nazis were more transparent than AI companies are today, but here we are. Truth doesn’t care what anyone thinks of it.

---

Here’s the sickening part. There are not enough books in the world to save AI.

There’s not. I’m sorry, but the math doesn’t work.

The massive datasets that powered earlier versions of AI are now contaminated. That’s the phrase they use in the AI industry. Contaminated. Wikipedia, blog sites, news media. They can’t scrape them anymore. Too much undisclosed AI.

They can’t risk undisclosed AI contaminating the model.

They need human text.

Physical books are the last refuge. The only place to get uncontaminated text, but there’s not enough. Because if they scan every book in the world, 130 million books published since the dawn of publishing, it would yield about 40-50 trillion tokens.

They need 100 trillion tokens to build the next generation of AI models.

So if they do scan and destroy every book ever published, they’ll still end up with only half the tokens they need. So where will the rest come from?

Let me tell you where.

Ingram Content Group, one of the biggest book distributors in the world, recently introduced an opt-out form for publishers who do not want their books to be purchased for AI model training. They said due to the increase of AI companies buying books to use for AI training, they are giving publishers an opt-out.

But they added a caveat.

They caution publishers that they’ll do “their best” to ensure that books aren’t sold to an AI company for scanning purposes but due to the increase in anonymous purchasing, they may not know the true source of every transaction.

Which means even if a publisher opts out?

Shrug. Maybe the books were bought by a middle man. How are they to know?

Every day, I see people saying AI is the future of writing. No it’s not. I am. Human writers, like me, who do not use AI to write or edit the work. I am the future of writing because AI cannot exist without us. And if we can figure out how to keep our writing out of the claws of AI, it will choke to death on it’s own probability and sameness.

But we don’t know how to do that, yet. Because middle men doing the work for a slice of the profit will sell any book to an AI company if they know the book is human.

And I can’t help but think of Death, in the opening of The Book Thief, saying he needs a vacation because he can’t bear to watch the leftover humans.

The survivors.

Left crumbling among jigsaw pieces of realization, despair and surprise.

Standing there with punctured hearts and beaten lungs.

But we’ll figure it out. We’re not there yet but we’ll get there because humans are smarter. AI didn’t invent itself, we did. AI can’t train on itself, it needs human work to survive. And when we figure out how to keep human work out of the hands of AI, we’ll see what the future looks like. I suspect it will look like humans making the kind of work that people using AI think they’re making while they pollute their own tool.

AI is not the future. Not as far as writing goes. Maybe for sorting data and spotting patterns, sure. But as far as creative writing goes, the future is human.

---

Here’s my AI policy. I don’t use it for writing. At all. That would be like asking it to eat cake for me. I’ve been writing for over 30 years, why would I outsource what I love doing? When AI can clean my damn kitchen, maybe we’ll talk.

---