Booksellers across the country are reporting something unprecedented: mysterious bulk orders for secondhand books, entire inventories vanishing into the void. The books arrive at warehouses, get scanned for their text, then get pulped. This is not a metaphor. This is what happens when an industry discovers that training artificial intelligence requires consuming the written word at industrial scale, and nobody bothered to ask whether destroying physical books to create digital ghosts was the move.
The secondhand book market, which existed for centuries as a refuge for readers too broke or too principled to buy new, has become a resource extraction operation. Thrift stores report their inventory depleting faster than donations arrive. Rare book dealers watch in horror as bulk buyers show up with spreadsheets and zero interest in the actual content. The irony is so thick it needs its own training dataset: the same tech industry that claims to democratize knowledge is liquidating the physical infrastructure that made knowledge accessible in the first place.
Why would AI companies do this instead of just licensing digital archives or working with publishers? Because bulk secondhand books are cheaper, unregulated, and come with plausible deniability. A company can claim the books are being “repurposed” or “archived” while they’re actually being shredded into training tokens. The market has responded by treating secondhand books like a commodity, which means prices are up, availability is down, and the entire ecosystem of little libraries and community book swaps is now competing against algorithms that don’t read but consume.
The absurdity isn’t that AI needs data. The absurdity is that we’ve decided the most efficient path to artificial intelligence is the systematic destruction of actual libraries.