Companies are purchasing physical books because books published before the widespread adoption of generative AI offer something...

2025-11-28 · Source: Pascal’s Substack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Emerging Technologies & Innovation · Depth: Intermediate, long

Summary

AI companies are actively acquiring and sometimes destroying physical books published before 2022 to secure high-quality, human-written training data untainted by generative AI output. This practice, facilitated by services like ISBNdb which helps clients acquire 1,000 to one million books per order, risks removing rare works from circulation and enclosing cultural knowledge within private datasets. The destruction, often for faster scanning, raises concerns about the loss of original artifacts and the transfer of publicly accessible material into opaque corporate systems. This trend highlights a growing scarcity of "clean" human data, potentially leading to "model collapse" from recursive training on synthetic content, higher costs for model improvement, and a concentration of AI development among firms with proprietary data access.

Key takeaway

For Directors of AI/ML evaluating data acquisition strategies, recognize that relying solely on bulk-purchased, destructively scanned books introduces significant ethical and performance risks. Your teams should prioritize transparent, licensed data sources with clear provenance to avoid "model collapse" and ensure long-term sustainability. Invest in non-destructive digitisation and quality assurance to preserve cultural assets, maintain public trust, and secure high-quality human knowledge.

Key insights

AI's reliance on human-generated data is creating a scarcity crisis, leading to unsustainable practices like destroying old books.

Principles

• Recursive training on synthetic data causes "model collapse."
• Human-created knowledge is AI's scarce natural resource.
• Data provenance and quality are critical for model performance.

Method

ISBNdb offers large-scale book acquisition services, using ISBNs to locate titles, avoid duplication, and organize destructive scanning for AI training.

In practice

• Implement transparent, non-destructive digitisation practices.
• License content directly from authors and publishers.
• Establish robust OCR quality control for scanned texts.

Topics

• AI Training Data
• Model Collapse
• Data Scarcity
• Generative AI Contamination
• Book Digitisation
• Data Ethics
• Cultural Preservation

Best for: Research Scientist, Investor, CTO, AI Scientist, AI Ethicist, Director of AI/ML

Related on AIssential

• The $200 Billion Data Debt: Every Major AI Lab Is Running Out of Fuel and Nobody Is Talking About… — The AI industry faces a critical data debt crisis as high-quality human-generated data is exhausted and synthetic data causes model collapse.
• AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop — Pre-2022 print books are a critical source of "clean" AI training data, free from model-degrading AI-generated content.
• The Next AI Revolution Won’t Be Bigger Models — It Will Be Better Data — The next AI revolution will prioritize better data quality and smarter learning processes over simply building larger models.
• The Irony of the AI Boom: Why Tech Giants Are Paying Humans to Teach Chatbots How to Write — Human writers are crucial for training advanced AI models, exposing a critical data quality bottleneck in the AI development race.
• AI vendors are destroying rare books to feed the chatbot — AI vendors are systematically destroying physical books for training data, utilizing legal rulings and specialized intermediaries.
• AI is deteriorating in realtime — AI models risk "model collapse" and increased hallucinations when recursively trained on synthetic data.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Pascal’s Substack.