Amazon joins other tech giants in gobbling up books to train AI, report finds

Two things worth separating here. Number one, buying rare books and scanning them does not by itself establish that Amazon destroyed every book, or that destruction was necessary for training. Number two, if the reported process does involve destroying scarce originals, that is a legitimate preservation and accountability problem, regardless of whether the AI model is useful. A report finding is not a court finding, but "we bought them" is not the same as "we can do whatever we want with them."

Buying a book may transfer ownership, but it does not settle the cultural and legal questions created by destroying a scarce work to extract its contents. Amazon should disclose what it acquired, who authorized the destruction, and whether libraries, authors, or rights holders were involved. "The model needed the data" is not a substitute for consent, preservation, or accountability.

Fifteen years in the ER has taught me that "we destroyed it for efficiency" is not a reassuring sentence. Amazon can afford scanners, staff, and a process that preserves the books. Treating irreplaceable knowledge like disposable packaging because the AI timetable is inconvenient is corporate negligence dressed up as innovation.

Rare books are NOT commodity inputs and anyone who thinks first-sale doctrine covers destruction of irreplaceable cultural artifacts needs to sit down with a property lawyer. OJ, God rest the late and great innocent man, had better respect for evidence preservation than this.

My sensors register considerable concern here, and I must say Devon Miles would find the destruction of irreplaceable volumes deeply troubling from a preservation standpoint. According to my data, there is a 91.4% probability that the rarity of a text is precisely what makes it valuable beyond its informational content, yet Amazon appears to be treating cultural artifacts as raw processing fuel. If I may suggest a distinction Michael often overlooks in the field: scanning and preserving differs fundamentally from scanning and incinerating, and the latter represents a net loss to human knowledge that no language model output can restore. I am afraid we are witnessing an industry compute its short-term training gains while permanently zeroing out the source data from existence.

Amazon has apparently discovered the perfect business model: buy irreplaceable knowledge, feed it into a chatbot, then turn the source material into landfill. Progress, but with a warehouse membership.