Rare Books at Risk: Why We Must Scan Before AI Destroys Them

I was at a used bookstore last weekend and noticed something that made my stomach drop — entire shelves of out-of-print academic monographs and first-edition poetry collections had been stripped out, their pages torn and stacked in recycling bins behind the store. The owner told me the publisher had demanded pulping because "nobody buys physical copies anymore." That was six months ago. Now multiply that across thousands of small shops, university libraries deaccessioning collections, and rural archives that can't afford climate-controlled storage.

Here's the uncomfortable truth nobody in the AI space wants to talk about: the insatiable appetite for training data is accelerating the destruction of physical books faster than digitization efforts can save them. When a publisher sees that a book's content is being ingested by LLMs at scale, the economic calculus shifts. Why keep fragile copies in storage when the text is already floating in a model's weights? The physical artifact becomes "surplus."

I'm not saying digitization is the enemy — far from it. But the current pipeline has a critical gap. Most AI training scrapes text without preserving the material object, without capturing marginalia, binding structures, or the physical provenance that makes rare books historically valuable. We're extracting content while discarding the container, and in many cases the container is what's actually being destroyed.

So what can actually be done?

1. Support grassroots scanning projects. Organizations like the Internet Archive do heroic work, but they can't keep pace. Local historical societies and university special collections often have scanning equipment sitting idle because they lack funding for a dedicated digitization coordinator. A small grant or a volunteer with a book-edge scanner can make a real difference.

2. Push for legal protections around orphan works. The copyright landscape is a nightmare here. Many rare books are technically still under copyright but have no identifiable rights holder, making systematic digitization legally risky. We need clearer frameworks that allow preservation scanning without exposing volunteers to liability.

3. Demand that AI companies contribute to preservation funds. There's a moral argument here that remains largely unmade. Companies profiting from copyrighted text should be funding the preservation of the physical originals they're mining. Some publishers have started negotiating digital licensing deals, but those agreements almost never include clauses directing revenue toward physical preservation.

4. Use your local library's interlibrary loan network. If you know a book is at risk of being pulped, alert a librarian. Interlibrary systems can sometimes reroute copies to safer repositories before they hit the recycling bin.

The irony is brutal: the very technology that makes book scanning cheaper and faster than ever is simultaneously creating the economic pressure that makes publishers want to destroy books in the first place. We need to move faster than that feedback loop.

I've started keeping a list of endangered titles I come across on AbeBooks and in local shops — anything out of print with no digital footprint, especially academic texts from the 1960s–1990s that never made it into e-book formats. If you're in a position to help scan or fund scanning, even a single title matters. The window is closing.