AI and The New Age of Book Burnings

Helping haul eighteen boxes of secondhand books, £17,000 worth, into the back of a UPS truck was one of my tasks at Plurabelle Books, the bookshop where I worked over the summer. Most had been on our shelves for years, accumulating a thick layer of dust, but were soon to be on their way to a warehouse in Chicago. Postal pickup day usually excites thoughts of whose fingers will soon turn the pages of the book you just packed for shipping, but these books were never going to be in the hands of anyone again. Their spines were to be sliced, and their pages efficiently scanned to train AI large language models. 

The phrase “destructive scanning” emerged from the court proceedings surrounding Anthropic’s Project Panama, an operation described as an “effort to destructively scan all the books in the world.” The process involves purchasing physical books, removing their bindings, scanning their pages, then shredding the printed originals, all in the interest of assembling a large body of high-quality, human-authored text with which to train their chatbot, Claude. For AI companies, books are a stable source of complex, long-form language that an internet increasingly saturated with AI-generated text cannot offer. The scanning of books, however, is not new. “This all started 20 years ago with Google,” explained Michael Cahn, owner of Plurabelle Books and my boss. By recruiting Tom Turvey, formerly Google Books’ head of partnerships, Anthropic took Google’s book-digitisation to a hyper-efficient extreme. Now, other AI companies are following in a  race towards a greater intelligence. Amazon, introduced as “Earth’s Biggest Bookstore,” predictably led to the closure of one-third of independent bookshops across America between 2006 and 2009. Amazon’s digitisation of the book trade, now paired with its “reduction of books to the information contained within them” for AI, as Cahn put it, corners bookshops into accepting these large sums of money and the associated moral burden. 

Back at my work desk was a first edition paperback of Ray Bradbury’s Fahrenheit 451, reserved by a collector and to be sold in person that coming Friday. Fahrenheit 451 follows Guy Montag, a book burner in a surveilled world where short-form media has monopolised information and where his growing doubt towards his job is one he lacks the words or power to describe. Destructive scanning has been likened to the book burnings that inspired Bradbury, where 25,000 books were thrown onto literary pyres by German university students during World War II. Whilst the Nazis toyed with fire’s dual symbolism of destruction and rejuvenation, it would be too on the nose for AI companies. Opting for the eco-fascist route, Anthropic pledged to recycle the pages of the books it has destroyed, all the while recycling their content to train AI. It is this “recycling” (or rejuvenation) that led District Judge William Alsup to find Anthropic within its rights to use books for training AI, given they were processed in a “transformative” way, thus avoiding copyright under “fair use”. 

Books, however, are destroyed every day. Cahn himself rummages through the UCLA library bins to save books destined for a recycling plant, only to bin them if they are ultimately too common a copy and not worth selling. As both a bibliophile and a bookseller, books are invaluable yet dispensable products. The problem is, therefore, not the destruction of books; it’s when that destruction becomes ideological. 

Book burnings follow a historical pattern, ordered by those in power to protect a doctrine and weaken those who stand against it. Just as dangerous as the total extinction of a book, however, is its preservation for a select few. The Roman emperors Augustus and Tiberius, for example, collected and burned thousands of prophetic, astrological, and political books. Those preserved were only to be consulted by a state-appointed committee of priests, allowing an emperor to manipulate prophecies that would legitimise his rule while making private fortune-telling a capital offense. 

Once again, books are being destroyed not to remove all trace of the information they hold but to centralise it. “If all knowledge is concentrated into one entity, that is surely a problem,” stated Cahn. When the companies responsible for the disappearance of physical books are also creating a machine to outsource intelligence, you don’t need a tin foil hat to see the path we’re on. We seem to be approaching Orwell’s dictum of “he who controls the present controls the past,” or at least he who controls the production, organisation, and circulation of knowledge controls the intellectual record of humanity, and thus our perception of who we are.

Illustration by Sandra Palazuelos Garcia