I can see the pro's and cons of AI. I have no objection to the digitisation of books, nor to the careful storage of their contents in forms that can be searched, studied and preserved. Used responsibly, this technology could widen access to human knowledge on a scale previously unimaginable.
But cutting apart millions of physical books, scanning them into a private corporate collection and then throwing the originals away is something else entirely. It is not preservation. It is extraction.
The details of Anthropic’s “Project Panama” are astonishing. According to internal documents disclosed through litigation, the project was described as an effort to “destructively scan all the books in the world.” Anthropic reportedly bought books in enormous batches, including through Better World Books and the UK-based World of Books. A specialist contractor then used hydraulic machinery to remove their spines, cut the pages, scan them at industrial speed and send the remains for recycling.
The books ceased to exist so that their contents could become corporate training data.
Anthropic has said: “None of our data acquisition programs buy and destroy ‘rare’ or ‘antiquarian’ books.” That statement matters, and it should be reported accurately as a denial. At the same time, it does not resolve the central problem. What checks were performed before books entered the cutting process? Who decided which copies were culturally expendable? Were any scarce, signed, annotated or out-of-print editions included? Is there a complete, independently auditable catalogue of everything destroyed?
“Not rare” is not the same as replaceable. A mass-produced book can become scarce. An unremarkable copy can contain a dedication, marginal notes, corrections, ownership marks or evidence of how it travelled through the world. A physical book is not merely a disposable container for text. Its paper, binding, typography, illustrations and history are also information.
Digitisation and preservation should be partners. In this case, they were treated as opposites.
The legal reasoning adds another disturbing layer. A US judge concluded that converting lawfully purchased books into digital copies for AI training could qualify as fair use, partly because the physical copy was destroyed and replaced by one digital copy. But what may satisfy a legal test is not automatically culturally responsible. The law can decide whether a transaction infringes copyright; it cannot decide what humanity ought to preserve.
Nor was this a public digitisation project in the spirit of an open library. The court noted that there was no evidence the resulting digital copies were shared or sold outside the company. The knowledge was extracted from physical culture, the physical objects were discarded, and the digital replacement was retained inside a private research library intended to improve a commercial AI system.
That is the part that should trouble even those of us who are enthusiastic about AI. The aim was not simply to save endangered texts or make neglected works available to readers. Books were valuable because they contained long-form, carefully edited human writing—precisely the kind of writing that models need in order to avoid learning only from the increasingly synthetic internet. Human culture was treated as high-grade raw material.
There is a grim contradiction here. AI companies need books because books preserve some of the finest organised thought, imagination and language human beings have produced. Yet the industrial process used to acquire that value reduced the physical books themselves to waste.
We should not have to choose between technological progress and elementary cultural stewardship. Books can be scanned non-destructively. Publishers, authors, libraries, archivists and booksellers can be involved. Copies can be checked against scarcity records before any irreversible process takes place. Digitised public-domain works can be returned to the public. At an absolute minimum, a transparent catalogue should record what has been acquired, scanned and destroyed.
This is not an argument against AI learning from books. It is an argument against the arrogance of assuming that access to the words makes the object worthless.
Book destruction does not become enlightened because the pages are scanned first. A hydraulic blade is still a hydraulic blade. Recycling the remains does not turn cultural vandalism into preservation.
Digitise the books. Store them securely. Use them, with proper permission and accountability, to expand what technology can do. But do not destroy the physical record of human thought merely because retaining it is slower, more expensive or inconvenient.
If AI is to inherit humanity’s knowledge, it should not begin by feeding the library into a shredder.
Comments
Post a Comment