AI Companies Purchasing Used Books in Bulk for Training Data

Used bookstores are seeing a significant increase in sales, with artificial intelligence companies emerging as major buyers of physical books. These companies are acquiring large volumes of literature to use as training data for their artificial intelligence models, a practice that has raised concerns about the potential destruction of books.

One example is Barter Books in the UK, which typically sells 2,000 to 3,000 books weekly. Recently, the store received a single-day order for the same quantity from a company based in Canada. Similar demand has been reported by other bookstores in the region, though the exact purpose of these bulk purchases remains largely undisclosed by the buyers.

The trend aligns with recent legal developments in the AI industry concerning copyright. Companies like Anthropic have faced lawsuits for using copyrighted material to train AI. However, court rulings have indicated that such training may be permissible if the companies compensate for the material used. Acquiring used books offers a more cost-effective and potentially less scrutinized method for AI firms to obtain training data compared to direct licensing from publishers, some of whom may refuse to supply AI companies.

Anthropic has previously described a process involving "destructive scanning" of books, where pages are scanned at high speed for AI ingestion, after which the physical pages are discarded. This has sparked worries among book enthusiasts and collectors that rare and antiquarian books, often found in used bookstores, could be inadvertently destroyed. While Anthropic has stated they do not acquire or destroy rare books, the broader practice raises ethical questions.

It is plausible that other AI companies are employing similar methods. Reports suggest that facilities, such as Amazon's AI training center in Las Vegas, are also using books for training, with some potentially rare volumes included in shipments. Although many sellers cannot identify the end buyers due to anonymizing platforms, the sheer volume of orders suggests a use beyond personal collections. For struggling bookstores, this influx of business provides a much-needed financial boost.