Independent bookstore owners across Canada are reporting a surge in unusal bulk orders for obscure and aged titles. These sellers suspect the books are being acquired to train artificial intelligence models before being discarded as waste.
Innodata and Zoom Books: The New Bulk Buyers
The trend has manifested through unexpected inquiries from companies that do not fit the profile of traditional book collectors or decorators. Peter Sellers, owner of the Toronto-based Sellers and Newel bookstore, reported receiving suspicious calls starting around June from buyers seeking random titles without the typical size or color requirements associated with film set staging. According to the report, Sellers was contacted via LinkedIn by a representative from Innodata, a New Jersey-based data engineering firm that claims to provide domain expertise for seven of the world's largest tech companies.
Similarly, Patrick Hempelmann of BMV Books in Toronto has seen a shift in his customer base. As reported, a company called Zoom Books, based in Pitt Meadows, British Columbia, began purchasing hundreds of volumes from him approximately six months ago. These transactions occurred primarily through the online marketplace AbeBooks, with the buyers showing little to no specific focus on a single subject, instead acquiring a wide array of unrelated titles.
Graduate Math and $30 Technical Texts
The specific nature of the books being targeted suggests a hunger for specialized, high-density information that may not be readily available in clean, digital formats. Patrick Hempelmann of BMV Books noted that Zoom Books focused heavily on specialized technical titles with small print runs, including graduate-level physics and math texts. These books, often priced between five and thirty dollars, had in some cases remained unsold on the shelves for a decade.
Other Canadian booksellers have noted a pattern of requests for old medical texts and obscure educational volumes. this preference for specialized, out-of-print material indicates that AI developers are looking for "edge case" knowledge—complex technical data that can refine the accuracy of large language models in professional or academic fields.
The High-Speed Scanning Pipeline for Large Language Models
This activity is part of a broader, more aggressive strategy by AI firms to secure training data as digital sources become depleted or legally protected. the report describes a process where major AI companies spend millions of dollars on physical books, remove the spines, and use high-speed scanners to digitize the pages. Once the text is ingested into the system, the physical remains of the books are often discarded.
This physical-to-digital pipeline represents a loophole in the current copyright struggle. By purchasing physical copies, companies may believe they are bypassing some of the digital licensing hurdles that have led to lawsuits between authors and AI developers. For bookstore owners like Peter Sellers, this creates an ethical dilemma: the ability to clear out unsellable inventory versus the knowledge that the books may be destroyed to feed a machine.
The Missing Proof in Pitt Meadows
Despite the strong suspicions of the bookstore community, a critical gap remains in the evidence. while Zoom Books describes itself as a reseller and recycler of secondhand books, there is currently no definitive proof that its bulk acquisitions are destined for AI scanning. The report notes that these suspicions have largely been fueled by online discussions and patterns of behavior rather than leaked corporate documents or whistleblower testimony.
Furthermore, the source primarily presents the perspective of the booksellers. There is no recorded response from Zoom Books or Innodata regarding these specific allegations of "scan-and-scrap" operations. It remains unclear exactly how many physical volumes are being processed this way and which specific AI models are benefiting from this Canadian paper harvest.
Comments 0