Used book sales are surging. Is AI to blame?

Recently, used bookstores around the world have experienced an unusual surge in large-scale book orders, which is believed to be part of the process of securing data for AI training.

A US court has ruled that the use of copyrighted books for training by AI companies such as Anthropic constitutes 'transformative use' and does not violate copyright law. During this process, it was revealed that companies internally used a method called 'destructive scanning', which involves disassembling book bindings to quickly scan pages and then recycling the remaining parts, under the name 'Project Panama'.

This phenomenon includes the following key issues:

• Data diversity: The value of book data has increased due to the need for rare or diverse topic texts to improve LLM performance.
• Ethical concerns: Concerns from booksellers and preservation experts regarding the physical destruction of rare or old books during the training process.
• Differences in legal regulation: While the US broadly recognizes 'transformative use', other countries like the UK require prior permission from copyright holders, leading to different legal interpretations by country.