For over a year, booksellers have raised concerns that companies in the AI sector are purchasing large quantities of rare books, only to destroy them after scanning for training purposes. This notion has been difficult to substantiate until now. A report by 404 Media reveals that an Airtag discreetly placed in a rare book confirmed the involvement of a major technology company, specifically Amazon, in these bulk acquisitions.
On Monday, 404 Media shared how it partnered with a bookseller to embed an Airtag in a rare book from a significant order. This tracking device subsequently led to an Amazon AI training facility situated in Las Vegas, where a team was reportedly engaged in disassembling books to scan their pages. The facility's branding, featuring a logo of a Tyrannosaurus rex poised to devour a book, underscored the company's apparent disregard for the growing criticism surrounding the practice of destructive book scanning.
In response to these revelations, Amazon chose not to address the specifics of 404 Media’s findings. Instead, the company reiterated its prior position, stating to Ars that it acquires books through commercial means to develop and refine its products and services, without explicitly mentioning AI training.
Nonetheless, Amazon is working on what it terms frontier AI models, which necessitate vast amounts of distinctive training data to maintain a competitive edge against industry leaders like Google, OpenAI, and Anthropic. Currently, many firms are protective of their training datasets to avoid compromising their advantages. The potential to train models using text from rare and hard-to-find books could provide Amazon with a significant benefit, particularly since competitors like Anthropic and xAI have openly declared their avoidance of utilizing rare or antique books for their training processes.
