Per a 404 Media report published this morning, Amazon, that evil empire, is buying massive quantities of books, scanning them for AI training data, and destroying them in the process.
This development arrives on the heels of slower, but persistent, disclosures that other AI players have engaged in destructive scanning to train their large language models for at least the last couple of years.
Things began to slide into view this January, when The Washington Post revealed that Anthropic was pursuing a covert project to destructively scan all the books in the world.
Details about this so-called “Project Panama”—a nod to a Bond-villain vibe—surfaced during a widely publicized copyright lawsuit authors filed against Anthropic earlier this year.
Although the leak painted the company behind Claude as uniquely monstrous, ongoing lawsuits against Meta, Google, Microsoft, and OpenAI now point to comparable practices.
Across the board, major corporations are appropriating authors’ work to train their LLMs, with a notable emphasis on texts from dead writers who cannot raise objections.
Amazon had previously denied engaging in “destructive scanning.” (Or at least, they’ve remained very, very quiet about it.) But as Emanuel Maiberg reported, fresh signals hint at possible involvement.
How can we assess the situation? In a straightforward piece of investigative work, 404Media tracked a rare book they “suspected would be acquired by an AI company for training data,” and followed its path across the country:
That final destination was an Amazon warehouse in Las Vegas, Nevada. Amazon employees who work at this location say all they do is receive massive shipments of printed books which they then cut the bindings off in order to scan the books more quickly. The printed book is destroyed in the process.
But there’s even more Bond-villainy behind the curtain! Apparently, the team’s logo features a dinosaur, brandishing its teeth and with a book in its hands.
So the Rex is out of the park, Mr. Bezos.
You may wonder: beyond the silent plaintiff aspect, what makes rare books so valuable for AI ventures?
According to documents WaPo unearthed in relation to Project Panama, one Anthropic co-founder theorized that training AI models on books could teach them how to write well instead of merely mimicking low-quality internet speech.
Maiberg echoed this line of thinking, noting that printed books are valuable training data because much of their text isn’t readily available on the internet, which AI companies have already scraped.
Printed volumes are also neatly organized, and if they were published before 2022, they are likely free of AI-generated text, a factor that could help prevent a detrimental recursive loop known as model collapse.
Here’s hoping for a positive outcome.
There is, however, a silver lining. For the most part, the American public remains broadly opposed to the destruction of books, at least on the optics front.
Some companies have already begun to respond to public pressure about this unsightly and overly blunt practice.
So, assuming they don’t implode under pressure, may the collective cloud of shame curb these developers and their machines.
