An investigation using a hidden Airtag has revealed that Amazon's VGT3 facility in Las Vegas is scanning and destroying printed books, including rare ones, to train its AI models. Booksellers are concerned about the ethical implications and potential loss of historical knowledge.
This investigation highlights a potentially destructive method used by major tech companies to acquire training data for advanced AI, raising ethical concerns about the preservation of cultural heritage and intellectual property.
An investigation has revealed that Amazon is scanning and destroying printed books, including rare and antique ones, to train its artificial intelligence models. The practice came to light when a bookseller, suspecting AI firms were bulk-ordering and destroying books after scanning, planted an Airtag in a rare book. The Airtag was tracked to Amazon's VGT3 facility in Las Vegas, where a team reportedly tears books apart to scan their pages.
Amazon has declined to comment specifically on AI training, stating only that it purchases books through commercial channels to develop and improve its products and services. However, the company is actively developing advanced AI models that require vast amounts of unique training data to remain competitive.
Discussions among workers at the VGT3 facility, as flagged by 404 Media, suggested that Amazon had previously faced a shortage of books for scanning, raising concerns about the facility's ongoing operations. Booksellers theorize that AI companies are systematically targeting books with ISBN numbers to ensure comprehensive coverage of unique works in their training datasets.
The practice has drawn criticism from book lovers and collectors who fear the loss of historical, intellectual, and sentimental value contained within these rare works. Some critics have labeled the destructive scanning as 'evil incarnate.' While some Redditors debate the necessity of preserving books that have sat unread for decades, others express concern that AI firms will hoard this knowledge, making only 'warped, censored, and paywalled fragments' of ideas available.
Pick the topics you care about. Get only what matters, on your cadence.