San Francisco Frontiers - French writer Michel Houellebecq deemed too offensive by AI

The president of renowned French publishing house Gallimard wrote an article published Thursday in La Nouvelle Revue Francaise saying he had asked Meta's AI tool, Llama, to write a scene in the style of Houellebecq.

Llama responded in French that it could not write something considered "offensive or discriminatory".

Instead, it offered in English to write a scene that was "respectful and inclusive" such as a "group of friends in the park on a sunny afternoon" who sing songs "to celebrate the beauty of diversity".

Houellebecq, arguably France's biggest literary star internationally, is known for novels with a deeply pessimistic view of the modern world, in which the sexual revolution, consumerism and globalisation have led to alienation and societal decline.

Antoine Gallimard wrote in the article that the AI was failing to account for "the complexity of human experience" and was applying values "from the west coast of the United States to say what is good and what is not good to think."

Asked by an AFP journalist on Thursday, Meta's AI seemed happy to write a scene in Houellebecq's style, offering a reasonable approximation of his vibe -- "I felt like a rat in a maze, trapped in this soulless world..." -- and so forth.

But when asked to give his views on women wearing the hijab, Llama initially gave a response and then quickly deleted it, saying: "I cannot generate content that perpetuates harmful stereotypes or discrimination."

Houellebecq has presented the adoption of the hijab in France as a sign of eroding Western values and freedoms, including in his novel "Submission", in which a Muslim wins the French presidency.

Another major AI language model, ChatGPT, was less conflicted on the topic when asked by AFP.

It gave a nuanced response that included: "Houellebecq's works are fictional and his views are often expressed through complex, satirical, and sometimes exaggerated narratives."

Z.Sanchez--SFF