AI Companies Buying and Destroying Old Books for Training Data

Recent reports and viral videos have sparked outrage over AI companies purchasing large quantities of old, out-of-print books to use as training data for their models, then destroying the physical copies after scanning. The practice, known as destructive scanning, was first reported by 404 Media in July, and later investigated by The Atlantic's Alex Reisner. The books, mostly non-fiction guides and manuals from the 1970s to 1990s, are bought in bulk from bookstores in Germany, the Netherlands, Canada, New Zealand, and other countries. The company behind ISBNdb, a bibliographic database, is said to have facilitated these orders, though it claims it was just a market test. While destroying books is legal and not strictly necessary, it is efficient for AI companies, as it allows high-speed scanning. The practice raises concerns about the destruction of cultural heritage, though most of the books are not rare and have little value. Legal experts note that recent US rulings consider this a format change, not copyright infringement, and using copyrighted material for AI training is deemed 'transformative.' However, fears persist that AI companies might target rare books in the future, and that concentrating human knowledge in AI systems could have negative consequences for authors and society.