Why AI companies are buying and destroying books, even rare and ancient ones: the assistance of Fair use in the USA

There history that you are about to read is so disturbing that it doesn’t seem real, yet it is yet another clear sign that we are getting ever closer to the dystopian future told in TV series and Hollywood blockbusters. Obviously it can only involve theartificial intelligence (AI)around which, evidently, the destiny of a large portion of humanity revolves. But let’s get to the point. As reported by 404 Media and The Telegraph, a British newspaper, for some time i booksellers – even those who deal with ancient books, rare and precious – are receiving orders with lists of hundreds of volumes to buy in bulk. They are sent by intermediary companies for the giants of the Silicon Valley that they need data to feed to your models machine learning. Books written before 2022 – therefore free of “contamination” from AI – are considered fundamental for training models. And not just books, but also mapsmanuals, documents, essays and any other text resulting exclusively fromingenuity and human intellect.

So far there would be nothing strange, given that being material legally for sale, it can be purchased by anyone, including women Big Tech. The problem is what happens to these books afterwards, too out of catalogue and maybe in single copy. In very simple words, to be scanned and used to feed the AI, they are deprived of binding, dismembered And cut from hydraulic machines which speed up the data acquisition process. But it doesn’t end here. If this is already enough to make the operation completely unacceptable, the next step is disturbing. The books are not returned to their original state, or at least “patched up”, but literally destroyed. Exactly, like i book burnings linked to dark and dark periods of ours history. As reported by The Thelegraph, there are services that allow AI companies to purchase used books in bulk, from 1,000 to 1 million volumes. And then there are the cases of antiquarian booksellers, who receive lists with requests for hundreds of volumeswhen customers generally purchase one or very few copies of what interests them. The British newspaper quotes the story of a Dutch bookseller from Haarlem who received a email with the request for hundreds of books listed in a spreadsheet. He thought it was a scam, but it was all true.

But why do AI giants need these texts? And above all, why do they destroy them after scanning them? The answer to the first question is simple: some have already tried it with pirated books in PDF which are found online but it went badly, given that after being “discovered” requests for them flooded in billions in compensation by authors and publishers, for non-payment of licenses and the like. But by purchasing huge quantities of physical books, companies take ownership of them and on the basis of first purchase doctrine they can do whatever they want with it, including scanning and destroying it. But why destroy them? Here, as explained by The Thelegraph and 404 Media, a controversial American law comes into play related to copyright. Simply put, the digital copy derived from the scan is considered “Fair use” if it passes as a transformative actbut not how redistributed copy. Fair use is a principle of US copyright which, in certain circumstances, allows you to use protected material without asking permission from the rights holder. If I scan a book to feed it to AI and then destroy it, according to AI companies I perform the act of transformation necessary for Fair use and I strengthen mine legal defense (a judge has already recognized this in a lawsuit). In essence, a sort of “quibble” is exploited in which the transformation occurs by making the copy an aggregate of data to be exploited to feed the AI ​​model. With the destruction of the book, in fact, one can appeal to Fair Use without having to pay the authors or ask for licenses. Clearly, when ancient, rare and out-of-print works are at stake, the destruction causes an enormous loss for thehumanity and the culture.

AI companies need large quantities of books written before the advent of AI because they are not contaminated from the latter, therefore they are works of pure human intellect and intellect. Paradoxically, the texts on web they are no longer considered valuable for powering the models precisely because the vast majority of them are now totally generated by AI, therefore, generally, rubbish unsuitable for training. And buying used books, dismembering them for scanning and destroying them is much less expensive than finding licensing deals with publishers and authors. As always, it’s all about money, interest and greed.