Key Points
- An AI startup co-founder says current AI is unlikely to make major scientific breakthroughs.
- He argues that AI models are designed to be agreeable and predict the most likely outcome, rather than being contrarian.
- This is the opposite of how great scientists, who often challenge the consensus, think.
- His view contrasts sharply with the hype from major AI leaders, such as Sam Altman.
Today’s artificial intelligence models, such as those that power ChatGPT, are unlikely to produce any major, Nobel Prize-level scientific breakthroughs. That’s the verdict from Thomas Wolf, co-founder of the AI startup Hugging Face, who is pushing back against some of the biggest hype in the tech world.
Wolf argues that the very way these AI models are designed makes them poor candidates for true, revolutionary discoveries. He points out two key problems. First, chatbots are built to be agreeable and align with the user’s prompts. Second, and more fundamentally, they are designed to predict the most likely next word in a sentence simply.
This, he says, is the exact opposite of how great scientists think. “The scientist is not trying to predict the most likely next word,” Wolf explained. “He’s trying to predict this very novel thing that’s actually surprisingly unlikely, but actually is true.” True scientific breakthroughs often come from contrarian thinking that questions the consensus, something current AI is not built to do.
Wolf’s comments are a direct challenge to the bold predictions of major AI figures, such as OpenAI’s Sam Altman and Anthropic’s Dario Amodei, who have suggested that AI could dramatically accelerate the pace of scientific discovery.
Instead of making discoveries on their own, Wolf believes these AI tools will be more like a “co-pilot for a scientist.” They’ll be excellent research assistants, helping humans sift through data and generate new ideas, but the spark of genius will still have to come from a human brain.
