A major online library is calling for volunteers to digitise and preserve physical books, warning that time is running out. The reason cited: AI companies are buying printed works, scanning them, and destroying them in the process, with the risk that knowledge ends up locked away on private servers.
On 6 August we ran an article on this phenomenon of buying and destroying printed books. Here is the reaction it prompted, and it deserves a closer look.
Recap of the mechanism
Why have books printed before 2022 suddenly become worth their weight in gold? Two reasons we detailed.
First, the web has filled up with generated text, making it hard to tell a human source apart. A printed book from before the arrival of mainstream generators is clean by virtue of its date. Second, fast industrial digitisation means cutting off the book's spine so the pages can pass through an automatic scanner. The book does not survive.
We had set out the important nuance: destroying a common copy is no tragedy, a novel printed in hundreds of thousands of copies does not disappear. The problem concerns rare editions, those that end up in a lot bought by volume with no regard for content.
What this initiative proposes
The call rests on a simple idea: if books are to be digitised, better they be digitised in a framework where the result remains accessible to all, rather than in a private corpus.
The central argument is about monopolisation. A book digitised by a company to train its model enriches that model. The digitisation benefits no one else, and if the physical copy has been destroyed in the process, public access has not been replaced, it has been removed.
This issue intersects with copyright, and it would be dishonest to present this initiative as a moral no-brainer. Digitising and distributing a protected work without authorisation remains a copyright infringement, whatever the intentions. This is precisely the debate we covered with the publishers' lawsuit against Google, and it is not settled. One can find it legitimate to want to preserve a heritage and still consider that doing so without a legal framework is a problem. Both positions are defensible.
What this episode reveals
Beyond the specific case, it illustrates a tension that runs through the whole question of training data.
On one side, a resource that has become rare and therefore valuable. On the other, a heritage that belonged to everyone and to no one, whose exploitation by private players raises a question of the common good.
It is the same logic we noted regarding the release of a scientific model by a state: when a resource becomes strategic, the question of who controls it ceases to be technical.
What to take away
This initiative will not solve the problem, and its promoters probably do not claim it will. Volunteers will not compete with industrial operations funded by labs with considerable resources.
What it does do, however, is make the issue visible. Three weeks ago, almost no one knew that books were being bought by the pallet and destroyed to feed corpora. The fact that this now prompts an organised response is itself information.
There remains the most uncomfortable question, and no one can answer it: how many editions that existed nowhere else have already disappeared under a scanner's blade, before anyone even thought to care?