<\/p>\n
\u201cI personally have mixed feelings about all of this,\u201d one small bookseller told tech news website 404 Media last week<\/a>. <\/p>\n The bookseller explained that in April, he suddenly went from selling around 20 books per week to hundreds \u2014 including obscure and out-of-print works.<\/p>\n \u201cIt benefits me financially as well as by clearing out old inventory that is otherwise unlikely to sell,\u201d he said. \u201cI\u2019ve been well suited for these sales with inventory from overseas and foreign language books. On the other hand, I don\u2019t like the end-use, and I don\u2019t like that uncommon books are being pulped.\u201d<\/p>\n Rare booksellers across Europe are seeing a similar surge in purchase requests which they suspect are coming from AI labs in America.<\/p>\n Pieter de Vries, an antiquarian bookseller in Haarlem, recently received an email from a person identifying herself as Nataly from Singapore-based company 2077AI, Dutch news site BNR reported<\/a>.<\/p>\n The woman wrote that the company was undertaking a \u201ca new project focused on collecting books in multiple languages, currently mainly in English\u201d. \u201cWe have compiled a very extensive list of editions that we are currently trying to acquire, and we plan to place a fairly large order,\u201d the email said.<\/p>\n The attached list contained 3000 English-language titles organised by ISBN number, ranging from books on fairytales and folklore to technical manuals and science texts.<\/p>\n Media outlets in the Netherlands, Switzerland, Spain, and Germany report booksellers have received nearly identical requests, per NL Times<\/a>.<\/p>\n Booksellers believe the purchase requests were unrelated to collecting or resale due to the obscure, highly specialised nature of the titles, concluding they were being used to train AI models.<\/p>\n Large language models (LLMs) that power AI tools like ChatGPT until now have mainly been trained on vast datasets that include internet data, licensed books and articles, code repositories and human feedback.<\/p>\n But with much of the content available online now exhausted \u2014 and increasingly polluted with poor-quality, AI-generated writing<\/a> \u2014 AI labs are turning to uncorrupted texts published pre-2022.<\/p>\n One company, ISBNdb, which boasts that it has the \u201cworld\u2019s largest book database\u201d, now offers bulk book buying for AI labs, \u201cup to one million titles per order\u201d, including of \u201colder, rare and specialist volumes\u201d.<\/p>\n \u201cThe world\u2019s best AI training data is sitting on a shelf,\u201d it states<\/a> on its website.<\/p>\n ISBNdb notes that \u201cprint books from the pre-LLM era are structurally guaranteed to be free of this contamination\u201d.<\/p>\n \u201cMillions of the most valuable books have never been digitised. They exist only in physical form, scattered across library shelves, used bookstores, and out-of-print catalogues. We get them to you at scale.\u201d<\/p>\n Anthropic, the Google-backed firm behind Claude, and Facebook and Instagram owner Meta, have previously been sued in US courts for downloading millions of pirated books to feed into their AI models.<\/p>\n Last week, a federal court in the US signed off<\/a> on Anthropic\u2019s landmark $US1.5 billion ($2.15 billion) settlement in a copyright class action brought by a group of authors who accused the company of misusing their books. It\u2019s one of dozens of similar cases brought against AI firms in the US, and the first to settle.<\/p>\n While that case was still unfolding, Anthropic was hatching a new plan to avoid similar copyright headaches, dubbed \u201cProject Panama\u201d. <\/p>\n Beginning in early 2024, the AI lab quietly launched<\/a> an industrial-scale book-buying program, partnering with used-book wholesalers to acquire millions of paper books by the pallet for what internal documents \u2014 which became public last year during the copyright lawsuit \u2014 described as \u201cdestructive scanning\u201d, where books\u2019 spines are cut apart so the pages can be scanned and then discarded.<\/p>\n Anthropic described the project as \u201cour effort to disruptively scan all the world\u2019s books\u201d in an internal memo, which noted \u201cwe do not want it to be known that we are pursuing this project\u201d.<\/p>\n Court documents detailed how the AI lab used a \u201chydraulic powered cutting machine\u201d to \u201cneatly cut\u201d millions of books before scanning the pages \u201con high speed, high quality, production level scanners\u201d.<\/p>\n The Anthropic copyright case set the precedent \u2014 the US court ruled that by making a digital copy and destroying the physical in a one-for-one transfer, the process is deemed \u201ctransformative\u201d and therefore protected by fair use.<\/p>\n \u201cAnthropic kept Project Panama confidential, not because it was illegal, but because it looked bad,\u201d ISBNdb says in an article<\/a> on its website. \u201cDestroying millions of books evokes images of burning libraries, even if the law was on their side.\u201d<\/p>\n ISBNdb, like other bulk buying services, promises to keep AI buyers anonymous, noting that \u201cthe optics problem is real\u201d. \u201c\u2018AI company destroys two million books\u2019 is not a headline that generates sympathy,\u201d it states, while arguing that \u201cthe framing of destruction as vandalism mistakes the nature of what is actually happening\u201d.<\/p>\n \u201cA physical book is a delivery mechanism for information,\u201d it continues.<\/p>\n \u201cOnce that information has been extracted and encoded into an AI model, the delivery mechanism has served its purpose. What remains is paper, ink, and binding material. The book is not destroyed. Its value has migrated. The paper returns to the material cycle; the knowledge enters the intellectual one.\u201d<\/p>\n SpaceX founder and rival AI developer Elon Musk, for his part, said he was opposed to the practice of destructive scanning.<\/p>\n \u201cI\u2019ve asked the SpaceXAI team to preserve any rare books in a library and scan them the hard way vs just cutting off the spine and scanning,\u201d he wrote on his social media site X.<\/p>\n Now Australian book dealers are urging Anthony Albanese to protect local titles from the practice, under the Prime Minister\u2019s newly announced Office for AI.<\/p>\n<\/p>\n