Secondhand book sales are booming. Is it because of AI?
Independent booksellers are reporting a surge in bulk purchases of secondhand books, often shipped to warehouses, leading to suspicions that AI companies are acquiring them for training. This trend is linked to a US court ruling that deemed using purchased books for AI tr…
Intelligence analysis by Gemini 2.5 Flash
A mysterious boom in secondhand book sales is puzzling independent booksellers globally, who are receiving unusually large orders for diverse titles. The prevailing theory is that these books are being acquired by AI firms for "destructive scanning" to train large language models, a practice that raises ethical concerns about the destruction of physical books, despite some booksellers…
Imagine a super-smart robot brain that learns by reading tons of books, like how you learn new words and facts. To teach it everything, some companies are buying huge piles of old books from shops, even really old ones. They then carefully take the books apart to quickly scan every page into a computer, like making a digital copy, and then recycle the paper. It's a bit sad for the books, but it helps the robot brain get super smart!
Analysis
The revelation of "Project Panama" within Anthropic's internal communications sheds light on the industrial scale at which AI companies are seeking to acquire and process textual data. This initiative, described as aiming to "destructively scan all the books in the world," involves a systematic process where physical books are shipped to specialized facilities. Here, their spines are removed to enable rapid, high-volume scanning of individual pages, after which the remaining paper is recycled. This method, while highly efficient for digitizing vast libraries, underscores the significant logistical and technological investment required to feed the data demands of advanced AI models.
Booksellers like Stuart Manley from Barter Books, who has observed unprecedented bulk orders, note the seemingly random and diverse nature of the titles being purchased. These range from obscure Latin texts to popular cowboy novels, suggesting that AI models benefit from an extremely broad and varied dataset rather than specific genres or authors. This indiscriminate acquisition strategy is driven by the need to expose large language models to the full spectrum of human language and knowledge, thereby enhancing their ability to understand, generate, and contextualize information across myriad topics. The mystery surrounding the ultimate destination of these books has fueled widespread discussion among booksellers, who are left to speculate on the precise mechanisms and ethical implications of these large-scale data acquisition efforts.
Judge William Alsup
A pivotal moment enabling this large-scale data acquisition was a US court ruling by Judge William Alsup, which significantly impacted the legal landscape for AI training. In a lawsuit brought against AI firm Anthropic by three authors, Judge Alsup determined that the company's use of purchased books for training its AI software was "exceedingly transformative." This legal interpretation is crucial because it allows AI companies in the US to digitize and utilize copyrighted material, provided the physical copies are legitimately acquired, without necessarily violating existing copyright laws. The ruling effectively provides a legal precedent for the "fair use" of copyrighted works in the context of AI model development.
This decision has far-reaching implications for both the intellectual property rights of authors and publishers, and the operational strategies of AI developers. While it offers a degree of legal certainty for AI firms operating within the US, it also highlights the global disparities in copyright legislation. Professor Emily Hudson of Oxford University points out that UK copyright laws, for instance, would typically require explicit permission from copyright owners for such extensive acts of copying and training. This divergence in legal frameworks creates a complex international environment for AI companies seeking to build comprehensive training datasets, potentially leading to different operational approaches and legal challenges depending on the jurisdiction.
McNaughtan's
The ethical dimension of destructive scanning is a central concern for independent booksellers, who are custodians of cultural heritage. Derek Walker, owner of Edinburgh bookshop McNaughtan's, articulates a nuanced perspective on the issue, distinguishing between the destruction of widely available academic texts and truly rare or unique historical editions. While the loss of a common book might be deemed acceptable, the destruction of an 18th-century edition, especially if it is the sole surviving copy, represents an irreplaceable loss to cultural history and scholarship. This distinction underscores the need for careful consideration and potentially more selective practices in data acquisition.
Despite these concerns, some booksellers, including Stuart Manley, acknowledge a practical benefit to the surge in sales. They note that many common titles, such as "The Da Vinci Code," are no longer in high demand and have sat on shelves for years. The opportunity to sell these unwanted books, even if they are destined for recycling after scanning, provides a welcome boost to trade and helps clear valuable shelf space. This creates a dilemma where economic pragmatism clashes with ethical considerations regarding the preservation of physical artifacts. The challenge lies in finding a balance that supports AI innovation while safeguarding irreplaceable literary and historical treasures.
Key points
- Independent booksellers are observing a mysterious surge in bulk purchases of secondhand books, often shipped to overseas warehouses.
- The prevailing suspicion is that AI companies are acquiring these books for "destructive scanning" to train large language models.
- A US court ruling deemed using purchased books for AI training as "exceedingly transformative" and not a copyright violation.
- Anthropic's internal "Project Panama" aimed to "destructively scan all the books in the world," involving removing book spines for rapid digitization.
- While some booksellers appreciate clearing old stock, there are ethical concerns about the destruction of potentially rare or historically significant books.
The boom in secondhand book sales provides an unexpected revenue stream for independent booksellers, helping them clear old stock that might otherwise go unsold for years. This process also contributes to the advancement of AI, potentially leading to more sophisticated and knowledgeable language models that can benefit society in various ways.
The practice of destructive scanning raises significant ethical concerns about the potential loss of unique or historically valuable physical books, which could be inadvertently destroyed in the pursuit of AI training data. Different copyright laws globally could also lead to legal complexities and disputes over intellectual property rights.
