Destruction by Definition
I know that AI labs are buying up books by the bulk load in order to feed more unique training text into their large language models. The reports of OpenAI and Anthropic engaging in this practice are more than a year old at this point, probably more. Ingesting physical books into their digital equivalent involves scanning each page, which can be subsequently converted to text via OCR. If you have a lot of books to scan, ripping out the pages will allow you to do this faster than manually turning and processing each bound page.
While I know all this, 404 Media’s recent report on Amazon’s VGT3 team struck me as a particularly dystopian sign of the times. Amazon is digitizing physical books, just like all the other labs. The VGT3 team simply executes this task. Their logo is a tyrannosaurus rex preparing to chomp on a book.
While I won’t shed a tear for the excess copies of books that get consumed in this process, many of the books destroyed are in limited supply. I would hope that we could slow down the process a bit and show some reverence for the art that might be disappeared by this process. It is perfectly possible to digitize a book without destroying it. While some of this may be the valuation of speed above all, combined with the casual disregard for external costs, I suspect the destruction is partially by design. If these rare books vanish, the owner of the digitized content will have a unique training input that could give it an edge over the competition. The collective commons of books is being replaced by a zero-sum game where the powerful fight over limited resources.
A massive maw driven to consume by a tiny reptilian brain captures the zeitgeist quite well. Perhaps a meteor shower would be a welcome change.
