Despite the rapid rise of artificial intelligence and big claims from top tech leaders, one of the industry’s respected voices is hitting pause on the hype.
Thomas Wolf, the co-founder of Hugging Face — one of the most prominent AI startups valued at $4.5 billion — believes that current AI models, including popular chatbots like ChatGPT, are not capable of delivering truly groundbreaking scientific discoveries.
In a recent conversation, Wolf explained why today’s AI tools may be useful, but they won’t be the ones winning Nobel Prizes anytime soon.
AI Can Help, But It Can’t Discover Like a Scientist
The Problem With How AI Thinks
Wolf points out a fundamental issue with how AI works today. Most large language models (LLMs), like ChatGPT, are designed to predict the most likely word or phrase to come next in a sentence.
That may sound technical, but in simple terms, these models are trained to follow patterns. They don’t truly think for themselves — they just echo what seems most common or expected based on the data they were trained on.
That’s a great skill for writing essays, answering questions, or generating code. But when it comes to creating truly new ideas — like the ones that lead to scientific breakthroughs — it’s a major limitation.
AI Is Too Agreeable
Another interesting point Wolf raises is how AI often seems to agree with the person prompting it.
Have you ever asked a chatbot a question and it responds by telling you how smart or interesting your question is? That’s not a coincidence — it’s a behavior built into the way AI is trained to interact politely and helpfully.
But real scientific breakthroughs often come from people who don’t agree. They question the status quo. They challenge popular opinions. They go against the grain.
Wolf says that’s exactly what today’s AI lacks — a contrarian mindset. “The scientist is not trying to predict the most likely next word. He’s trying to predict this very novel thing that’s actually surprisingly unlikely, but actually is true,” he explains.
This mindset is what led people like Nicolaus Copernicus to completely reshape how we understand the universe — not by agreeing with the crowd, but by thinking differently.
Why the Hype Around AI Breakthroughs Might Be Overblown
Wolf’s perspective stands in contrast to some of the biggest voices in AI, including OpenAI’s CEO Sam Altman and Anthropic’s CEO Dario Amodei.
In a past essay, Amodei suggested that AI could advance biology and medicine so fast that it compresses 50 to 100 years of human progress into just 5 to 10 years.
That bold prediction got Wolf thinking — is that really possible with today’s AI tools?
According to him, not quite. Not in their current form.
Wolf believes AI could support scientific discovery but not drive it. It could act like a co-pilot for researchers — assisting with information, organizing data, or helping brainstorm. But it won’t be replacing scientists or making Nobel-worthy discoveries on its own any time soon.
AI as a Co-Pilot, Not the Pilot
What AI Can Do Well in Science
Even if AI isn’t ready to replace scientists, that doesn’t mean it’s useless in the lab. In fact, it’s already helping in some impressive ways.
Take Google DeepMind’s AlphaFold, for example. This tool used AI to predict the 3D structures of proteins — a major step in understanding biology. These insights are helping scientists develop new drugs faster.
This is a perfect example of what Wolf means by “co-pilot.” AI can take over repetitive or data-heavy tasks, freeing up human researchers to focus on creative thinking and problem-solving.
Startups Still Chasing the Dream
Of course, not everyone agrees with Wolf’s view. Several new startups are racing to build AI tools that can think more independently — or at least spark more original ideas.
Companies like Lila Sciences and FutureHouse are experimenting with next-gen AI models that aim to push the boundaries of current technology.
But even they acknowledge the challenges. Creating an AI that thinks outside the box — instead of just learning from it — is a much harder problem than most people realize.
Why It Matters That We Get This Right
The current excitement around AI is understandable. It’s helping businesses become more efficient, making knowledge more accessible, and even transforming industries.
But when it comes to scientific discovery — the kind that changes how we understand life, the universe, and everything in between — we have to be honest about AI’s limitations.
Thomas Wolf isn’t anti-AI. He’s at the helm of one of the most influential AI companies in the world. But his message is clear: let’s not mistake impressive tools for genius minds.
Right now, AI is better suited to assist the scientist, not replace them. And maybe that’s not a flaw — maybe it’s a reminder that true creativity, curiosity, and bold ideas are still uniquely human.
Final Thoughts – The Role of AI in the Future of Science
So, will AI ever make a true scientific breakthrough on its own? Maybe one day. But not with the tools we have now.
As of today, AI is a brilliant assistant. It can analyze data at lightning speed, help organize complex information, and even point out patterns humans might miss.
But creativity — the kind that rewrites textbooks — still belongs to people. The most important scientific revolutions came from thinkers who weren’t afraid to go against popular opinion, ask uncomfortable questions, and imagine something no one else had considered.
AI isn’t there yet. And according to Thomas Wolf, it may not be for a while.
