Gluttony and the guillotine

Leave a comment
books

If you live in the continental United States and get bitten by a bat, or a raccoon, or a fox, your doctor will probably advise you to go and get rabies shots. Rabies isn’t common, but it’s common enough that about 100,000 people go through the preventative protocol every year.

This process is no joke. First you do a PCR test, then you get injected with immunoglobin at every site where the skin is punctured, and then you undergo a series of four vaccine shots over the course of two weeks. Side effects of these drugs include soreness, headaches, nausea and dizziness.

While this might sound bad, it’s a lot less bad than catching rabies, which starts by giving you flu-like symptoms and a tingling sensation, before swelling the brain, making you terrified of water, and causing explosive anxiety and confusion until it completely overtakes your nervous system and paralyzes you until you cannot breathe. The fatality rate for those who display clinical symptoms is 100%.

These rabies protocols help you avoid certain death, but there’s somebody for whom it’s even worse: the beast that bit you. If the little critter that bit you has rabies, it will definitely die. But if you catch a little critter that’s suspected of carrying the disease, it will be taken away to get tested—which involves euthanizing it before examining its organs and biochemistry for evidence. Thousands of animals are tested this way for rabies every year in the US, but only about 5% of those that are killed actually turn out to have had the disease. Rabies is fatal; testing animals for rabies is fatal.

It’s a paradox: To save the thing, you have to destroy the thing.

The latest controversy in AI is the revelation that Anthropic bought millions of old books, chopped them up, and then scanned the pages in to its systems. Large Language Models have an insatiable appetite for words, and this scheme—codenamed Project Panama, according to court documents—was seen as the best way to slurp in high-quality information into the maw of Claude. Insiders knew this would be controversial.

“Project Panama is our effort to destructively scan all the books in the world,” one Anthropic staffer wrote in an internal planning document, as reported by the Washington Post. “We don’t want it to be known that we are working on this.”

Partly it was just brute force, enacted under the influence of the scale drug. Why not devour books by the truckload?

It was also an end-run around the information ecosystem, deliberately cutting out authors and publishers by scooping up old titles; paying pennies for information that otherwise had to be acquired through more expensive means and agreements. And while the documents about Project Panama are from a while back, it’s not clear that this feeding frenzy has actually stopped. Whether it’s Anthropic or not, booksellers around the world are still reporting a puzzling surge in sales of hard-to-find titles.

“Is it just me, or has there been an uptick in the number of AutoBuy orders since the tail end of last year?” one bookseller wrote, according to 404 Media. “Any comment on what is happening? Is an AI going to read every single book? Any insight into how the selections are made? They seem to vary quite a bit in condition, format (hardcover and softcover), price and so on.”

In the end, though, the thing that really seemed to offend people wasn’t the machine’s hunger for words, or how the plan deliberately sidestepped intellectual rights; folks were most outraged by the actual, physical act of taking the book to pieces in order to scan it. The spine, sliced off by a huge guillotine; the pages, separated. The book, pulped. In Anthropic’s mind, saving the thing, requires destroying the thing.

Ripping books apart is only one of the never-ending cavalcade of reasons to dislike AI. (Please choose from a menu that includes unfairly profiting from the stolen work of artists and creators; the devastating environmental impacts; the sweeping layoffs driven by AI-pilled CEOs; the rogue cyberattacks enabled by security models; the overinflated financial investments made by desperate capitalists. They keep adding more, too: Just in the last few days, we’ve had the boggle-eyed claims that we should parent our children using ChatGPT-generated podcasts and that “we are now in the Singularity.” ) If you weren’t already skeptical of this new technology, its fans and proponents are hell bent on giving you more material.

But destroying books seemed to catch a certain wave. It was a sort of desecration, an offense against the soul.

“I personally have mixed feelings about all of this,” the bookseller, who suspects he’s sold hundreds of books to AI companies for training data, told me. “It benefits me financially as well as by clearing out old inventory that is otherwise unlikely to sell. I’ve been well-suited for these sales with inventory from overseas and foreign language books. On the other hand, I don’t like the end-use, and I don’t like that uncommon books are being pulped.”

And all of this was spiced up, of course, by Anthropic’s mustache-twiddling secrecy. We all love a good villain.

Here’s the thing, though. Books get destroyed all the time.

One of the sad and underreported facts about the publishing industry is how terrifically wasteful it is. Publishers will often demand that unsold copies are pulped, shredded, or ripped apart. Even in preservation efforts, books are often torn up since paper is a fragile medium.

Historically, books were not always seen as valuable. Expensive books have routinely gone uncared for, and been mistreated more often than dealt with reverently. Samuel Johnson used his copy of Shakespeare’s first folio as a coffee coaster. Dylan Thomas mislaid his manuscript of Under Milk Wood in a pub in London and nearly lost it forever. Byron’s memoirs didn’t even make it to the printed page, burned in the fireplace by his friends who deemed their contents too scandalous.

Today, carnage is even part of the business model and destruction is baked in to the balance sheet. Publishers and distributors routinely destroy books, ripping off their covers, pulping them, or otherwise rendering them unusable for all manner of reasons that don’t always make much sense from the outside.

It was deeply upsetting to me the first time I realized that hundreds of copies of a magazine we had made and lavishly paid to be printed would end up being pulped by the distributor if they weren’t sold. The contract we had to sign to get them to send our work to bookstores obliged them to commit this barbarous act; they could not return them to me, they could not send them elsewhere, they had to be simply destroyed. It felt like punishment.

While it is a perverse financial logic at play, this attitude is also a recognition that books are just objects. They are not holy relics. They are produced en masse, distributed widely, let out into the world to roam wild. I love books, but I do not treat them particularly nicely. When we read we consume, in more ways than one. We crack the spine, we make notes in the margins, we turn the pages until they fall out, we keep them in the sun until they yellow and break down.

There are good reasons we prize books. They represent knowledge. Banning them is a measure of intolerance, and burning them is a visible symptom of authoritarianism. But that is what they represent, not what they are. The information and the physical form are related but different.

Looking at it this way, Anthropic’s plan seems to be less fascistic than simply a business finding new ways to exploit a resource. After all, as far as I can tell, the company isn’t destroying “valuable” books, or even particularly old ones, nor is it disrespecting knowledge. Most of the books it is ingesting are out of print copies of textbooks, or work that’s unavailable in digital formats. It’s writing that is “correct” and untainted by the touch of AI—which is a delicious flavor combination if you’re an LLM with a garbage in, garbage out problem.

All that said, the fundamental premise of Anthropic’s approach is wrong.

You don’t have to destroy the thing to save the thing. You choose to do it.

You don’t have to kill animals to see if they’ve got rabies; it’s just cheaper and easier and less risky than running non-lethal tests. And you don’t have to destroy books to scan them. I remember when Google began its mass scanning project 20 years ago, which it had codenamed “Project Ocean” and announced at the Frankfurt Book Fair. People were freaked out at the prospect of millions of books suddenly being digitized, and I remember interviewing the head of the Bodleian Library—the great Oxford institution which was Google’s unlikely partner—and being stonewalled about the nature of the actual scanning system.

It turned out later that the company had developed a patented system to scan the books without destroying them, using infrared light to and measure the angle of the pages so that the words could be rendered inside the computer without the book’s spine having to be removed.

Before Google came on the scene, book scanning was a tedious process that sometimes resulted in the death of a book. The software used to scan books, called Optical Character Recognition software or OCR for short, required each page of the book to be flat. Now anyone who's ever opened a book knows it's next to impossible for a book to lie flat without some sort of device. One solution to the problem was to use glass plates that individually flattened each page, but this method wasn't very efficient. The other solution was to chop off the book's binding, but that method destroyed the book. How was one to go about scanning a book quickly and efficiently without destroying it?

This was two decades ago, proof that your machine can devour words without also making hard to find books a little bit harder to find.

Perhaps that’s one major difference. Anthropic’s most egregious offense may be that it has kept the results private, or at least filtered them through Claude rather than creating a library of scanned material that the public can access. They might not be burning books, but they are hiding them and profiting from them. And that is a move that’s worth a little frothing at the mouth.

Leave a comment