Amazon Scans Rare Books for AI Training, Reports 404 Media

Amazon has been acquiring a significant number of rare books, removing their spines, and scanning them to utilize the content for artificial intelligence training, according to an investigation by 404 Media. The report followed the placement of a tracking device in a rare book that arrived at an Amazon facility located in Las Vegas, known as VGT3, which features a logo depicting a dinosaur holding a book.

In a statement to 404 Media, Amazon acknowledged its activities by stating that it “purchases books through commercial channels to improve the products and services customers use.” The company’s need for large volumes of text data to enhance its language models has been underscored as it exhausts available material from online sources, with some organizations reportedly resorting to illegal methods to obtain content.

Rare books, particularly those that are out of print or difficult to find online, provide valuable new sources of training data for language learning models (LLMs). These texts are especially significant because content published prior to 2022 is assuredly original, authored by humans rather than generated by LLMs. Training on AI-generated texts, conversely, poses the risk of “model collapse,” a phenomenon where the output quality of an LLM deteriorates after too much exposure to AI-generated material.