Recent discussions on social media and in news articles have focused on allegations that AI companies are purchasing and destroying physical books, including rare ones, to train their models. Critics, including investor Michael Burry and entrepreneur Elon Musk, have expressed concerns about the implications of such practices. An article from 404 Media highlighted a surge in orders from booksellers, suggesting that rare books might be included, although it did not provide direct evidence linking any specific company to these purchases.
While it is known that at least one major AI company, Anthropic, has used physical books for training, the specifics of the current book purchases remain unclear. Reports indicate that many of the books being bought are not necessarily rare or valuable, but rather nonfiction titles published from the 1970s to the 1990s. Bookseller Charlie Becker noted that a significant portion of his recent sales went to a single buyer, but the nature of these books raises questions about their intended use.
The company ISBNdb, which was mentioned in the context of bulk buying for AI companies, has denied any involvement in purchasing or destroying books. Other companies, such as Zoom Books, have been implicated in large orders, but they also claim no connection to book destruction. Additionally, a Singapore-based organization, 2077AI, has reached out to booksellers for academic titles, further complicating the narrative.
Despite the lack of clarity, the notion of AI companies potentially destroying books has sparked significant public concern, drawing parallels to historical instances of book destruction linked to authoritarian regimes. The ongoing debate raises questions about the future of authorship, knowledge creation, and access to information in the age of AI.