AI Companies Destroying Rare Books to Train LLMs, Sparking Cultural Erasure Concerns

A report by Frank Landymore in Futurism reveals that AI companies, including Anthropic, are bulk purchasing out-of-print books, digitizing them, and then pulping the physical copies to train large language models (LLMs). This process, called 'destructive scanning,' is used to circumvent copyright laws. ISBNdb, a company supplying books to tech firms, offers bulk orders ranging from one thousand to one million volumes. The books are sliced by a hydraulic cutting machine, scanned, and then pulped. The practice is driven by the need to train LLMs on pre-2022 writing uncontaminated by AI-generated content. This has raised concerns about the destruction of rare and few remaining copies of books. The article draws a historical parallel to the burning of Michael Servetus's 'The Restoration of Christianity' in 1553, noting that only three copies survived, one saved by John Calvin. It also mentions that publishers destroy nearly 25% of works released each year due to poor sales. The piece criticizes the indifference of AI companies, quoting OpenAI CEO Sam Altman: 'We see a future where intelligence is a utility, and people buy it from us on a meter.' The author describes this as an act of 'libricide' and a war against the materiality of books, echoing George Orwell's warning that 'who controls the past controls the future.'