The AI industry’s appetite for high-quality training data has a grim side effect: physical books are being destroyed. The cheapest way to digitize millions of titles at speed is to chop off their spines, feed the pages through scanners, and toss the remains. For anyone who’s ever held a fragile first edition or marveled at centuries-old marginalia, the practice lands somewhere between vandalism and cultural erasure.
What makes the destruction harder to stomach is that a non-destructive alternative has existed for over 15 years. Google patented a scanning technology back in 2009 that could digitize books without tearing them apart. But the method isn’t perfect. Studies have documented issues with page curvature distorting text and pages getting missed entirely. Wired reported that when workers move too quickly, glitches occur—including disembodied hands accidentally obscuring the page during capture.
Those trade-offs matter to companies racing to train next-generation models. The Google approach is slower and more expensive, which makes it a tough sell when the goal is scanning millions of titles as cheaply as possible. And for rare books, it may not be the right tool anyway. The Internet Archive has spent decades demonstrating that proper preservation requires patience, careful handling, and human attention—things that don’t scale well when you’re building a training corpus measured in terabytes.
So the uncomfortable truth is that AI firms aren’t destroying books because they have no choice. They’re doing it because it’s faster and cheaper, and because the cultural cost doesn’t show up on a balance sheet. Unless preservation-minded institutions or regulators force a change, the economics will keep pointing toward the wood chipper.
