Is Amazon destroying rare books to train its AI models?
What's the story
Amazon has been accused of purchasing and destroying rare books in order to train its artificial intelligence (AI) models. According to an investigation by 404 Media, the e-commerce giant has been buying large quantities of these texts, cutting off their spines and scanning them for data. The practice was discovered when a tracking device placed in a rare book showed it had reached an Amazon facility in Las Vegas.
Training methods
Amazon defends its practices
Amazon has defended its practice, saying it "purchases books through commercial channels to improve the products and services customers use."
This is part of a larger trend where tech companies are using rare texts as a new source of training data for their AI models.
These books are especially valuable because they weren't written by an LLM before 2022, avoiding the risk of "model collapse" that can happen when too much AI-generated text is ingested.
Facility operations
The scanning process and its implications
The Amazon facility in question, VGT3, is represented by an icon of a dinosaur holding a book in its claws.
Workers at this facility have said they know little about the operation. Their main job is to receive large shipments of printed books and process them for quicker scanning.
While this may seem harmless, it can lead to the destruction of these texts in the end.
Strategic shift
Legal battles over AI training data usage
The book scanning issue gained attention after a lawsuit from authors against Anthropic, which revealed a program called Project Panama.
Under this project, books were acquired from commercial marketplaces, their spines were cut off, and they were scanned for AI training.
A judge ruled the practice as fair use, but it drew heavy backlash from authors, activists, and e-reader communities.
Meanwhile, Amazon has also revamped its AI strategy by discontinuing some in-house models.