AI companies destroy books, empowered by laws protecting corporations over people

In a North Las Vegas Amazon warehouse – designated LAS8 – sits a unit named VGT3, dedicated to buying rare books, shredding their spines and scanning them for AI training data. 

AI companies, striving for quality training data and restrained by the illegality of pirating, have resorted to buying up millions of used books. The spines are removed from these publications before they are fed into a massive scanner. Then the loose pages are recycled. The worst part is — this race to destroy our history in the name of machine innovation is completely legal.

For example, in 2024, AI company Anthropic, the parent company of UNLV’s partner AI platform Claude, started Project Panama, which internal company documents described as “our effort to destructively scan all the books in the world.” This is a project they admitted to hiding from the public. Within a year they had systematically shredded and scanned over 2 million books, all of which were scrapped afterwards.

To clarify, destroying books is not necessary to scan them. However, gutting a book of its binding, feeding it into a machine and scanning the loose pages is faster and less costly than manually scanning the pages one at a time. It is sheer greed that motivates this destruction.

The mad dash to scan every ISBN number in the world is founded on a lack of quality training data. What AI companies want is for their models to produce coherent and eloquent responses to user queries, but these bots are only as good as their training data. They cannot produce a single unique sentence on their own. 

The cruel irony is that AI programmers have run into trouble finding quality training data from the internet due to the proliferation of “AI slop.” This slop – low-quality AI content – has swarmed the internet and infected AI training data sets.

If an AI program is continuously trained on slop, it will produce more of it. AI companies felt they needed books to teach their agents not to mimic “low-quality internet speak,” as one Anthropic co-founder described it.

AI corporations ultimately decided the highest quality training data could be found in books. The irony is, they specifically sought out books written before 2022 to ensure they had no AI generated content themselves. This reflects the hellish nightmare we live in today, where AI has permeated both the internet and print media. 

So why are AI companies going through all the trouble of scanning books they could pirate online? Before embarking on a quest to tear apart every book ever written, AI companies en masse digitally pirated millions of books to train their AI agents with. Immediately, authors began to sue as their copyrighted materials were being used without their permission. This resulted in billion dollar settlements for copyright violation in cases like Bartz v. Anthropic.

The judge in Bartz v. Anthropic ruled that scanning and ravaging physical books for AI training data was considered fair use for being “quintessentially transformative.” This created a secure opening for AI companies to scan and shred physical books, as long as they were purchased, without fear of legal repercussions.

Fair use is the legal use of unlicensed copyrighted material for specific purposes — whether that be for education, entertainment, criticism, etc. Key factors courts use to decide whether something is an example of fair use are the purpose and character of the secondary use.

In recent decades, this factor has focused on the transformativeness of the secondary use. In other words — is the new use innovative? 

The fair use decision in Bartz v. Anthropic characterized the training of AI agents on books as transformative — as the original intent of a book is to entertain and/or inform the reader — not to train an AI agent. 

A human simply reading a book and learning how to better communicate information is by no means transformative — as that was the original intention of the publication. So it makes no sense that when an artificial intelligence improves its language capacity by analyzing a book, it is considered a completely new use.

Anthropic’s lawyers successfully argued that because a human learns differently than artificial intelligence, the use is transformative. This is utterly ridiculous. While a human was the original intended reader, and learns differently than artificial intelligence does, the end result — learning how to write better — is the same. This secondary use is not transformative.

What AI companies are doing — destroying rare books, some with no remaining copies — should be illegal.

It is time the courts stop protecting and enabling corporations over people, especially if the law is to be considered the ultimate equalizer. 

The only transformation AI companies are doing is transforming books into paper scraps. And they’re doing it right here in town.

Recent Articles

Related Stories

LEAVE A REPLY

Please enter your comment!
Please enter your name here