In the ongoing debate surrounding Anthropic's controversial practices, one question stands out: Why is a leading AI company choosing to destroy books? This intriguing query forms the basis of our exploration today.
The Bookish Enigma
Anthropic's 'Project Panama' has sparked a legal battle, with the company's internal documents revealing a surprising strategy. The aim? To 'destructively scan all the books in the world.' This raises a host of questions. Why opt for such an extreme method when securing copyright permission is an option? And why the secrecy?
A Matter of Training
The answer lies in the training of Anthropic's large language model, Claude. Claude needs an extensive dataset of high-quality language combinations, and books provide an ideal source for this. Anthropic's co-founder and CEO acknowledged the 'legal/practice/business slog' of managing copyright, which led them to initially use pirated sources. However, even this approach was deemed too complex, leading Anthropic to destructive scanning.
Legal Loopholes and Logistics
The US copyright law's 'fair use' doctrine allows for 'transformative' use of copyrighted works without permission. Anthropic's process involved physically destroying the books after scanning, a practice the court deemed legal, comparing it to teaching a human how to read and write. The logistics were intricate, with Anthropic hiring staff, managing warehouses, and even slicing the books' spines for easier scanning.
Cultural Significance and Regulation
The case highlights the cultural significance of books and the lack of legal provisions to protect them. The US, unlike many countries, has few regulations governing the treatment of books as cultural heritage. This raises concerns about the potential consequences if other AI companies follow Anthropic's lead. How many warehouses of destroyed books are acceptable? And what about the survival of rare and unique printed works?
Preserving Human Authorship
The 2022 threshold, when AI-generated text began to significantly impact the textual record, prompts a crucial question. Should we consider the 'wholly human author' as a category for collections preservation? Anthropic's actions suggest a devaluation of human authorship and creativity. We must ask ourselves why complex, long-form, uncorrupted text is so profitable and valuable to AI companies.
A Call to Action
The case of Bartz v Anthropic PBC serves as a powerful reminder of the importance of books and readers. It highlights the risk of ceding control to large language models and their proprietors. We must refuse to devalue our roles as creators and claim ownership of the cultural spaces where our thoughts and words thrive. The future of our large language lives depends on it.
