It was either that or writing. First praised and then reviled, Thélyson Orélien’s novel, “It Was Either That or Die”, has become a scapegoat for chatbots.
In light of the controversy, no writer can invoke Arthur Rimbaud’s line “I is another” when the poet no longer has control over what is expressed through him.
The use of a generative artificial intelligence (AI) solution must be disclosed. Morality demands it. The law requires it. Since August 2, the European regulation on artificial intelligence, the AI Act, has imposed transparency obligations. The use of this technology must be clearly indicated whenever there is a risk of confusion with reality, whether it involves the generation of text or images or interaction with a conversational assistant. It must be noted that few such disclosures have appeared, even in highly enthusiastic posts about AI on social media. It is then up to each individual to read them, or not, with full knowledge of the facts.
Public condemnation. In fact, as we pointed out here, “human becomes suspect.” Doubt is taking root. Whether justified or not, the use of a machine casts a shadow of disgrace over Thélyson Orélien. The book finds itself in the pillory based on a questionable verification and a message from an anonymous group.
Publishers and media outlets must grapple with AI, which is putting their bottom lines at risk at a time when no one seems to be reading anymore
Yet, to date, no solution for detecting AI-generated text is mature enough to claim complete effectiveness. There are far too many biases. It is particularly difficult to identify supposedly artificial text structures from authors whose writings have often, unwittingly, contributed to the training of language models. The snake bites its own tail and then hides in the grass. Thus, many texts written before the advent of generative AI are detected as being partially the work of machines.
Beyond that, it’s quite surprising to note that an app like Pangram, the one used to scrutinize the novel that has since been removed from the Prix Goncourt shortlist, sells the ability to “humanize” one’s text, no less. That is where its business model lies, and this has received little attention. It’s essentially both judge and jury.
Publishers and media outlets must come to terms with AI, which is helping to sustain their economies at a time when no one seems to be reading anymore. The very same people who point the finger at its uses have inevitably asked themselves how they might use it, if they haven’t already, to correct or translate texts, analyze manuscripts, and adapt works for other media, for example, by using synthetic voices.
Faced with a dizzying technological leap, industries must reinvent themselves, and professionals must take a chance on innovation, as they have often done before.
Faced with the explosion of “generated” content, the floodgates cannot hold. Flagging artificial content is a good start, but it is not enough. To give ourselves the best chance of maintaining a verifiable narrative, attributing original content, and promoting it to better monetize it, we will have no choice but to mark reality at the source. Much like carbon-14 dating, invisible and indelible solutions make it possible to authenticate and trace images and videos. For text, similarity technologies will make it possible to detect, based on just a few words, any fraudulent use that might be made of it. If the act of writing demands truth, technology can become the trusted third party.
This article was originally published in French in l’Opinion
This text, written in French by a human, was machine-translated into English and proofread by a human