Amazon destroys antique books to train its AI models
The e-commerce giant uses rare and out-of-catalogue items to train AI, sacrificing cultural preservation in the name of technological innovation
Amazon started out with books, but is now destroying them on an industrial scale to fuel the new technological frontier of artificial intelligence. The e-commerce giant, founded in 1994 as an online bookshop and later expanding into all product sectors, is buying tonnes of second-hand books, including rare volumes, to scan them and feed them into artificial intelligence systems to train language models.
The investigation by the 404 Media website
To make this process easier and quicker, they cut through the spine, effectively destroying the volumes. This was uncovered by an investigation carried out by the website 404 Media, which placed a tracking device inside a rare book to understand the purpose behind this trade in antique and rare books.
The suspicion was indeed that they were being used to train companies’ AI models, but it came as a huge surprise when the book’s journey across the United States led to an Amazon fulfilment centre in Las Vegas, Nevada.
The system used for these AI training operations is identified by the VGT3 brand and an image of a dinosaur clutching a book in its claws. Amazon stated in a press release that it “purchases books through commercial channels to improve the products and services used by customers”.
Books to feed language models
It is no secret that large language models such as ChatGPT, Claude or Gemini – Nova being Amazon’s proprietary model – require vast amounts of information and text to grow and improve their analytical and predictive capabilities. Having already drawn extensively from the internet, ancient and rare books – particularly those out of print or impossible to find – are now being used as a new source of training data.
