Why are AI firms bulk buying and pulping books? Because it’s cheaper than paying writers

Kenny’s, an Irish bookshop and online retailer based in Galway, has reported receiving unusual book orders, often placed via email during the night and requesting large quantities of obscure titles, sometimes exceeding 5,000 volumes. These orders align with a growing trend observed globally, wherein artificial intelligence firms are acquiring and physically processing vast numbers of printed books to feed into the algorithms that power chatbots and other AI systems. This practice, referred to within the industry as “destructive scanning,” involves slicing off the spines of thousands of books at once, digitizing the pages, and then disposing of the materials entirely. The act of pulping books has sparked unease among readers, who view it as a violation of the cultural and intellectual significance of written works. Books are traditionally regarded as vessels of knowledge, artistry, and historical continuity, serving as bridges between generations and repositories of human thought. The notion of machines consuming and discarding literary heritage evokes images of both overt censorship and more insidious forms of erosion, particularly concerning the livelihoods of writers and the integrity of creative expression. Internal communications from AI firms reveal a deliberate effort to conceal such activities. A memo from an executive at Anthropic, one of the leading developers of AI chatbots like Claude, stated that “Project Panama is our effort to destructively scan all the books in the world.” This initiative, shrouded in secrecy, highlights the extent to which some companies are willing to go to secure data for training their models. Much of the public awareness surrounding these practices stems from legal proceedings, notably a recent lawsuit filed against Anthropic in the United States. The case was initiated by a coalition of authors who alleged that the company violated their intellectual property rights by using their books without permission to train the Claude model. According to court documents, Anthropic initially attempted to acquire millions of books by downloading copies from pirated websites, which already deprive writers of rightful compensation. However, as the company faced increasing legal scrutiny, it shifted strategies and began scanning and destroying physical copies of books instead. This transition underscores the broader economic pressures driving the industry, where the cost of obtaining content through legal channels is deemed too high compared to the financial returns generated by AI services. The legal judgment highlighted two primary concerns raised by the plaintiffs. First, Anthropic had the opportunity to negotiate with publishers and pay fair prices for the rights to use the books, yet opted not to do so. The court noted that users and corporate clients of Claude contribute to over $1 billion in annual revenue, suggesting that the company prioritized profit over ethical considerations. Second, the integration of scanned books into AI models risks devaluing the work of human writers. While the court acknowledged that Claude did not produce exact copies or substantial imitations, it noted that AI-generated content could enable less skilled writers to compete effectively in markets previously dominated by professional authors. Recent research indicates that the proliferation of AI-generated content is altering the landscape of digital publishing. The number of e-books published monthly on Amazon has increased threefold since the launch of ChatGPT, according to studies. However, the quality of these publications, as measured by user reviews and other metrics, has declined. This trend has given rise to a new sector of online entrepreneurs offering quick-profit schemes, enabling individuals to generate and sell optimized content through automated means. Journalist Kashmir Hill recently uncovered an instance where an AI produced a biography of her, further illustrating the rapid expansion of this phenomenon.