Booksellers suspect AI firms destroying rare books for training data

👁 115 reads

Booksellers suspect AI companies are buying rare books, destroying them, and scanning the loose pages as the cheapest way to build large training datasets, according to reporting by Ars Technica. Book lovers fear the practice is happening on a greater scale than currently reported, with some physical copies lost forever.

The core concern is economic. Tearing out pages, cropping, scanning, and discarding them is faster and cheaper than careful handling - and AI companies are competing to train models on high-quality long-form text found in books. Google patented a non-destructive scanning method in 2009 that could offer an alternative, but studies have found that page curvature can distort text, pages can be missed, and workers moving quickly cause further errors. The cost and speed trade-offs may not appeal to firms scanning millions of titles.

The Internet Archive, which has employed book scanner Eliza Zhang since 2010, takes a different approach. The organisation tested commercial scanners featuring vacuum-powered page-turning arms but found they damaged brittle and rare volumes. Zhang uses a foot pedal to raise the scanner glass, adjusts cameras for each page, and tracks fold-outs so bonus materials are not lost. As the Archive noted, “clean, dry human hands are the best way to turn pages.”

// read moreHidden AirTag Tracks Rare Books to Amazon AI Facility in VegasNewsNvidia Takes Minority Stake in Data Center Firm CloverleafMagazineGoogle’s Gemini Notebook Can Now Pull Content From Purchased BooksAI Tools

Kinetyra