Amazon, the company that built its initial fortune selling books online, is now purchasing rare texts and destroying them to train artificial intelligence models. According to investigative reporting, the e-commerce giant acquires these books through commercial channels, removes their spines, and scans the contents at facilities including a Las Vegas location marked by a dinosaur logo. Amazon stated that these purchases are made to enhance its products and services.
The practice reflects a growing demand for training data among AI companies developing large language models. Standard internet sources have largely been exhausted as training material, prompting companies to seek alternative text sources. Rare and out-of-print books are particularly valuable for this purpose, as they contain authentic human-written content unlikely to have been generated by artificial intelligence systems.
This approach addresses a significant concern in AI development known as model collapse, which occurs when language models trained on AI-generated text experience degraded output quality. By sourcing pre-2022 publications, companies can ensure their training data comes exclusively from human authors, avoiding this potential pitfall. However, the large-scale destruction of rare literary works raises questions about cultural preservation in the age of artificial intelligence advancement.
Originally reported by Techcrunch →
About the Author
