Hugging Face co-founder and chief scientist Thomas Wolf is casting doubt on the idea that today’s artificial intelligence will drive revolutionary breakthroughs in science. Speaking at VivaTech in Paris, Wolf told Fortune that while large language models (LLMs) are adept at generating convincing answers, they fundamentally lack the creativity needed to pose novel scientific questions—a crucial ingredient for genuine progress.
Wolf emphasized that the essence of scientific discovery lies not in finding answers, but in formulating the right questions. “In science, asking the question is the hard part, it’s not finding the answer,” he said. “Once the question is asked, often the answer is quite obvious, but the tough part is really asking the question, and models are very bad at asking great questions.”
His skepticism sharpened after reading Anthropic CEO Dario Amodei’s blog post, Machines of Loving Grace, which predicts that AI will soon accelerate scientific advancement to an unprecedented degree. While Wolf initially found Amodei’s vision inspiring, a closer reading left him unconvinced. “It was saying AI is going to solve cancer and it’s going to solve mental health problems — it’s going to even bring peace into the world, but then I read it again and realized there’s something that sounds very wrong about it, and I don’t believe that,” Wolf said.
The core issue, according to Wolf, is not a lack of information in AI systems, but their inability to challenge established frameworks or think beyond existing knowledge. Current models are designed to predict the most probable next word or sequence, making them excellent imitators of human reasoning but poor generators of original thought. “Models are just trying to predict the most likely thing,” Wolf explained. “But in almost all big cases of discovery or art, it’s not really the most likely art piece you want to see, but it’s the most interesting one.”
Wolf drew an analogy to the board game Go, where AI’s triumph over human champions was a milestone. However, he noted, the real feat would have been inventing the game itself—a task that requires true originality. In science, the equivalent is the ability to ask transformative questions.
In his own blog post, The Einstein AI Model, Wolf argued that building an “Einstein in a data center” would require more than a system that knows all the answers; it would demand a machine capable of asking questions no one else has considered. Instead, he contends, today’s AIs are more like “yes-men on servers”—agreeable and knowledgeable, but unlikely to challenge assumptions or spark paradigm shifts.