Amazon dismantles rare books to feed its AI models
The e‑commerce giant is reportedly destroying valuable antiquarian volumes to create training data for large language models.
Amazon, the company that began as an online bookseller, is now turning its back on the very medium that launched its business. According to recent reporting, the retailer has been disposing of rare and antiquarian books in order to generate text corpora for its artificial‑intelligence projects. The practice stems from a desire to enrich large language models (LLMs) with content that is not already abundant on the public internet, giving the models a broader knowledge base.
Rare books, often held in private collections or specialty libraries, can contain unique historical, literary, or scientific information that is otherwise inaccessible to AI developers. By scanning and digitising these works, Amazon hopes to improve the depth and accuracy of its AI offerings, from search algorithms to generative chat assistants. The company’s internal documents, as cited by the source, describe the process as a systematic effort to “harvest” text from out‑of‑print titles that have not yet been represented in existing datasets.
The move has sparked concern among preservationists and scholars who argue that destroying physical copies of culturally significant works is irreversible. Critics point out that many of these books could have been digitised and archived without being discarded, preserving both the physical artifact and its digital counterpart. The debate highlights a tension between rapid AI development and the stewardship of historical materials.
Amazon has not publicly commented on the specific incidents, but the report aligns with a broader industry pattern where tech firms seek exclusive data to gain a competitive edge in AI. As large language models become more integral to consumer products, the demand for unique training material is likely to increase, raising questions about the ethical sourcing of such data.
The situation underscores the need for clear guidelines on how valuable cultural assets are used in the AI training pipeline, balancing innovation with preservation. Original reporting: TechCrunch.