AI challenged to draft a 21st-century Constitution

When constitutional scholar Dominique Rousseau asked four artificial intelligences to write a Constitution for the 21st century, he didn't do it out of laziness or to impress his law students: he wanted to present them with a real challenge. The provocative question was this: Can artificial intelligence match human intelligence when it comes to devising the fundamental rules of a society? The 22 Sorbonne students spent six months drafting their Constitution, then compared it with those produced by ChatGPT, Le Chat, DeepSeek, and Gemini. And here comes the twist: everyone feared that AI would be better, faster, and more creative. Instead, the real surprise was the regulatory poverty of the Constitutions written by AI. ChatGPT, Le Chat, and DeepSeek produced texts consisting of ten concise, generic articles, while the advanced version of Gemini managed 68 articles across 20 pages, but only thanks to very detailed instructions provided by the students themselves. Meanwhile, the Constitution written by the students spanned 61 pages and 197 articles, replete with nuances and difficult choices. Yanis Khellafi, one of the students, put it bluntly: “AI can be original, but it’s not creative. It's an aid, not a substitute for the underlying work.” This means that AI is excellent at shuffling around what it finds, copying and combining existing ideas, but it cannot produce anything that goes beyond a sum of familiar pieces. When it comes to conceiving new principles or balancing opposing values, the machine grinds to a halt and resorts to vagueness. Here we see the difference between writing simple rules and truly building a society: the latter requires experience, doubt, and compromise. One detail is worth reflecting on: the students used AI as an opponent, but also as a tool. “I couldn't let them be unaware of what will be their work tool in the coming years,” said Rousseau. There is a lesson here: technology does not replace the effort of human reasoning; it enriches it or challenges it. But the core of creativity remains ours. Here is a perspective that is usually overlooked: we think that AI threatens our uniqueness, but perhaps the real risk is that we stop exercising the muscle of doubt, stop engaging in dialogue, and stop getting our hands dirty with ambiguity. If we leave it to AI to decide what is right and what is wrong, we risk losing the ability to shape the future—not because the machine is too good, but because we gave up too soon. Creativity is not the sum of everything that already exists; it is the ability to go beyond what is. If you want this idea to become part of your way of thinking, on Lara Notes you can indicate it with I'm In — choose whether it intrigues you, whether you've experienced it, or whether it's a deeply held belief for you. And if this story turns into a conversation with someone—perhaps at the dinner table, perhaps in the classroom—on Lara Notes, you can capture that moment with Shared Offline: it's your way of saying that the conversation was valuable. This was from Le Monde, and compared to the 24-minute original article, you’ve saved over 20 minutes.