Home / Technology / AI Devours Rare Books: A Cultural Erasure?
AI Devours Rare Books: A Cultural Erasure?
2 Aug
Summary
- AI companies target pre-2022 books for cleaner training data.
- Scanning processes destroy original book copies for digitization.
- Rare volumes vanishing irreversibly raise cultural loss concerns.
AI companies are actively seeking out and processing physical books published before 2022 as primary training material. These older works are favored because they are considered to contain original, human-authored content unaffected by the proliferation of AI-generated text and potential data-poisoning techniques.
Data brokers like ISBNdb are reportedly supplying these books in bulk to AI developers. The scanning process often necessitates the destruction of the original book, with spines cut and pages fed through high-speed machines to accelerate digitization. This method, while efficient, is irreversible and has led to significant criticism.
Concerns are mounting as many of these books are exceedingly rare, with very few surviving copies that have already endured historical hardships. Unlike digital content or widely available modern publications, these scarce historical volumes cannot be easily reproduced once their physical copies are lost forever.
A recent U.S. court ruling on fair use suggested that destroying one legal copy for scanning to create a training dataset could be permissible. However, this ruling and the general practice are drawing parallels to major historical acts of cultural destruction.
