Artificial intelligence (AI) cannot write books, at least not good ones. Book lovers have so far agreed on this. AI is still surprisingly bad at creating suspense, say literary scholars. However, work is being done on this, as a case from France suggests.
There the novel “C’était ça ou mourir” received the Fnac literary prize. But now he has been removed from the shortlist for the Prix Goncourt, the country's most important literary prize, because, according to the French Writers' Association, there is a suspicion that the Canadian-Haitian author Thélyson Orélien had AI take a lot of notes.
The case still has to be clarified, so far it is a suspicion, the writer speaks of defamation. Regardless, AI has been causing excitement in the industry for a long time. On the one hand, it fascinates book lovers and publishers, as the 100 million euro investment by C.H.Beck-Verlag in the legal AI Noxtua shows. It is the largest investment to date in the publishing house's 260-year history.
But the industry is horrified by the chutzpah with which large AI providers have, without asking, used books as free raw materials in recent years to train their models so that they can even write sensible texts. Pirating, scraping, scanning – everything was used. Lawyers for the authors rightly say: The big AI developers are behaving like teenagers in the Napster era of the 2000s - only worse. Pay for content? Do not feel like.
So far, AI companies have largely escaped unscathed. Anthropic paid $1.5 billion in a settlement last year after downloading 500,000 books from illegal shadow libraries. Basically, they stick to their stance that books can be used as (almost) free raw material for training AI.
There was talk of the “biggest robbery in history” at previous book fairs. But there was also a certain level of shock in this country. They looked at America, where 145 trials are now underway. Among the most important are the lawsuits filed by the New York Times and the US authors union Authors Guild against Open AI. Recently, there have been more American newspaper publishers who suspect that their articles are simply being "scraped" - even if they are behind a paywall - and later used almost verbatim in bits and pieces to answer user queries.
The state of shock in Germany has now subsided. In March, the world's largest book publishing group, Penguin Random House, also filed a lawsuit against Open AI in Munich. The Carlsen publishing house followed in August. That's a good thing.
The first successes are evident: the music collecting society GEMA has achieved important stage victories in Germany that could have an impact on the book industry. In both industries there is an accusation that the AI also “memorises” the content, i.e. de facto saves it and ultimately returns it to users, at least in bits and pieces.
The authors shouldn't be happy too soon. Once you have learned it, it is almost impossible to unlearn it. And what happens if US courts rule completely differently than European courts? The big models are trained in America. But there the AI providers hide behind the “fair use doctrine”. As long as books were purchased legally, they could also be used for AI training, is their central argument, because AI creates “transformative new things”.
So far they've gotten away with it, even though an appeals court recently ruled against the fair use doctrine. The rule comes from the pre-AI era. And it makes a difference whether something is done for scientific or commercial purposes. Training may have originally been experimental and scientific, but it is now a billion-dollar business.
Trump's government has already sided with the AI providers: AI training with books should generally not be treated as a copyright infringement, the US Department of Justice informed the court at the beginning of September that will decide on the class action against Open AI. America has a strong interest in AI, “which sets standards worldwide”. Or in Trump’s words: “Whoever wins AI wins.” One can only hope that the judges are not influenced by this.
