Quietly Expensive Lessons in Library Etiquette

There is a certain, very specific dignity to the act of borrowing a book. One enters a quiet room, exchanges a polite nod with a person who has dedicated their life to the alphabetical arrangement of human thought, and promises to return the volume before it begins to accumulate fines. It is a system built entirely on trust, low-key social pressure, and the occasional threat of a fifty-pence penalty.

It is somewhat jarring, therefore, to contrast this gentle ritual with the recent activities of Anthropic, an artificial intelligence laboratory of considerable repute and even more considerable funding. On Monday, a federal judge in California put the final signature on a $1.5 billion settlement, bringing to a close a class-action lawsuit brought by a group of understandably perturbed authors and publishers. The sum is, by any reasonable standard, rather large. It represents a payment of approximately $3,000 for each of the estimated 500,000 books that Anthropic’s digital assistants digested during their formative years.

For the average novelist, whose usual financial trajectory involves a great deal of hope and a very small royalty check, a retroactive sale of $3,000 for a book that may have been out of print since the late nineties is a remarkably generous transaction. It is, in fact, probably the most efficient sales campaign they have ever run, even if they had to hire a team of class-action lawyers to complete the transaction.

The true elegance of the situation, however, lies not in the size of the check, but in the exquisite legal distinction that necessitated it. The court, in its infinite and highly structured wisdom, did not actually find that training an artificial intelligence on copyrighted text is illegal. Indeed, the judge ruled that the act of reading a book to learn how sentences work is a perfectly acceptable form of "fair use." In the eyes of the law, a machine may read as much as it likes, provided it does so with the pure, unblemished motive of self-improvement.

The difficulty arose entirely from the question of where the machine obtained its reading material.

Had Anthropic sent a very large team of interns to the local library with library cards and a flatbed scanner, the entire enterprise would have been perfectly legal, if somewhat exhausting for the interns. Alternatively, had they purchased half a million physical copies from a high street bookseller, they would have been hailed as the greatest patrons of literature since the Medicis, and would have received several very nice thank-you notes from the Society of Authors.

Instead, they chose to download the books from pirate websites with names like Library Genesis and Pirate Library Mirror. These are digital repositories that do not, as a rule, pay royalties, nor do they employ librarians who frown at you if you speak too loudly.

Thus, the $1.5 billion fine is not a penalty for intellectual theft, but rather a very expensive lesson in library etiquette. The law has essentially declared that you are entirely welcome to learn from the wisdom of the ages, but you must not obtain that wisdom from a gentleman in a trench coat behind the bicycle sheds. It is a distinction that feels less like a triumph of modern jurisprudence and more like a rule from a particularly strict Edwardian boarding school.

I am reminded of an afternoon spent in the company of a retired archivist in Shropshire, who once spent three weeks tracking down a missing copy of a local parish register, only to discover it was being used to prop up a wobbly table in the church vestry. "The information was perfectly safe," he told me, over a cup of lukewarm tea, "but the posture was entirely wrong."

Anthropic’s posture was, indeed, entirely wrong. By settling the case, they have avoided the indignity of a public trial, where executives would have had to explain to a jury of ordinary citizens why they believed "Pirate Library Mirror" was a legitimate commercial distributor of contemporary fiction. They have also, rather cleverly, prevented the case from reaching an appeals court. This means the ruling remains a single, isolated decision, leaving the rest of the AI industry in a state of delightful legal ambiguity.

Other technology giants, such as Google and OpenAI, are currently facing their own copyright battles. They will no doubt watch the Anthropic settlement with a mixture of relief and mild anxiety. The message is clear: the digital universe may be infinite, but the receipts must still be kept in a neat folder.

For the authors, the settlement is a bittersweet victory. They have established that their work has value, even if that value was calculated retroactively by a mathematical formula in a California courtroom. One wonders what the machines themselves made of the half-million books they consumed. Did they appreciate the subtle irony of a Victorian triple-decker novel, or did they simply catalog the adjectives for future deployment in marketing copy for a new brand of organic soap?

A friend of mine, who writes experimental poetry that is read by approximately twelve people and a very confused spaniel, recently discovered that his entire corpus had been ingested by a major language model. He attempted to converse with the model about his work, only to be told that the poetry was "unusually dense and highly resistant to summarization." He was, I believe, quietly flattered.

In the end, we are left with a landscape where the boundaries of human creativity are being redrawn by the rules of acquisition. The great library of the future is being built not with stone and mortar, but with algorithms and settlements. And as we move forward into this brave new world of automated intellect, we would do well to remember that while the machine may have a mind of its own, it still needs to show its library card at the door.