AI Companies Destroy Millions of Books for Training Data

AI companies are destroying millions of physical books to create training datasets for chatbots. This practice involves acquiring books anonymously through intermediaries, with a focus on titles published before 2023, which are considered valuable for providing high-quality human-written data. The demand for these books has significantly impacted the used-book market, with one bookseller reporting weekly sales increasing from about 20 to several hundred due to AI buyers. While this has been financially beneficial, concerns have been raised about the permanent loss of uncommon and out-of-print books. This practice is similar to Anthropic's 'Project Panama,' which involved destructive scanning of millions of books. Recent court rulings have deemed the scanning of legally purchased books into digital copies as transformative fair use, despite the destruction of originals. However, a federal judge recently approved a $1.5 billion copyright settlement against Anthropic for using pirated works to train its AI. In light of the backlash, some AI developers, including Elon Musk, have advocated for preserving rare books instead of destroying them.