AI tech giants buy books in bulk. Even rare books. Why? To destroy them!
It’s unbelievable but true. Amazon and Anthropic— are reportedly bulk-buying books, scanning them for AI training data, and then destroying them with giant shredders. Amazon, the largest online seller of books, buying books to destroy them is indeed ironic.
By Ajith Pillai in Chennai
In today's dystopian times, the truth often feels stranger than science fiction.
Those familiar with Ray Bradbury’s 1953 novel, Fahrenheit 451, will remember its unsettling tale about a future where a totalitarian government employs the fire service to seek out and burn the most illegal possession of all: books. The authorities believe that the printed word is the source of all conflict and unhappiness and that a population deprived of reading is better off than one that engages with literature and reflects.
Now Bradbury’s story has emerged in a real-life avatar with the revelation that at least two tech giants—Amazon and Anthropic— are reportedly bulk-buying books, scanning them for AI training data, and then destroying them by putting them through giant shredders.
On August 17, an investigation published by 404 Media, a journalist-founded digital platform, provided concrete evidence that Amazon engages in bulk buying of rare books, scans them for AI training data, and then destroys them.
The investigation by 404 Media journalist Emanuel Maiberg involved placing an Apple tracking device on an order of 1,000 books that Amazon had made with a rare book supplier in California. The consignment was tracked from California to an Amazon facility in Las Vegas known as LAS8.
At the northern end of the facility was a unit called VGT3, with a logo at the entrance featuring a T Rex holding an open book. It is here that the spines of books are removed, the pages are scanned to feed data to AI models, and the books are later shredded.
Maiberg’s investigations, as well as other media reports, have stirred up a controversy among book lovers. To quote an August 17 report in Forbes: “Authors, archivists, historians and bibliophiles have launched an online war against tech giants and artificial intelligence developers after an increasing mountain of evidence shows they’re buying rare and antique books en masse and feeding them to large language models—only to slice off their spines and destroy the original print copies in the process.”
The shipment of 1000 books tracked by 404 Media is only the tip of the iceberg. Forbes points out that Amazon is “buying massive quantities of books, scanning them for AI training data, and destroying them in the process.”
Amazon, the largest online seller of books, buying books to destroy them is indeed ironic.
Publishers are known to shred old stock, but they are not known to buy books to destroy them.
Anthropic, a big name in the AI space, is guilty of buying millions of books in the last two years, handing them over to contractors to pull apart their bindings, scan the pages in high-speed scanners and later destroy the scanned pages. This was revealed in a lawsuit (Bartz v Anthropic) filed earlier this year by writers who objected to the digital funeral of their books and to copyright infringement.
Anthropic's buy-scan-and-shred operation even had a code name within the company: Project Panama. One of the court documents, an internal company memo, quoted by The Guardian, explains what the project is all about: “Project Panama is our effort to destructively scan all the books in the world… Why use a codename? … Because we don’t want it to be known that we are working on this.”
Interestingly, no AI company wants new books to scan and shred. Their interest lies in books published before 2022, the year generative AI and large language models (LLMs) came into focus. In other words, they are keen on books written exclusively by humans, before AI.
As Kathryn James of the Lillian Goldman Law Library, Yale, noted in her column in The Guardian: “Anthropic needed data, lots of it, of very high quality. Books, as it happens, remain one of the best sources for complex, high-quality, long-form text.”
Older books, it is widely agreed, are not only better written but also more original, as they are born of human effort and thinking. They are also free of AI slop (inferior quality digital content generated with the help of artificial intelligence).
Anthropic can take credit for popularising the destructive scanning of books, which other AI labs are now following. As far as the tech companies are concerned, what makes it attractive is that it is perfectly legal under US law.
In fact, the First Sale Doctrine of the Copyright Act makes it legal for the buyer of a book to destroy it if he/she pleases. And creating a digital file within a workspace or company for internal use is also legal and does not infringe copyright. Court judgments have ruled that training LLMs on data scanned from purchased books is legit. AI labs have been showing physical ownership of books to do as they please without paying any royalties to the authors.
Prominent hedge fund manager and investor Michael Burry has been one of the harshest critics of destructive scanning. He has been quoted as saying that the practice of sacrificing rare books was “evil incarnate.” Elon Musk, who often likes to take a contrary position to other tech bros for effect, has stated that he has asked those training X-AI models to preserve rare books they scan in a library.
But what happens when AI labs have scanned all 130 million books written purely by humans before generative AI came into the picture?
Even that data will not be enough to feed the next AI model.
So, what then?
Will the labs have to depend on AI-generated copy or human text contaminated by machine learning?
Well, to get the real stuff, they will have to persuade humans to write and think again without AI!!
Ajith Pillai is Co-Editor, independentink.in