Why current AI models won't make scientific breakthroughs, according to a top tech exec

A leading figure in the tech industry has expressed skepticism about the potential of current artificial intelligence models to drive significant scientific advancements. Thomas Wolf, co-founder of the AI startup Hugging Face, suggests that prevailing AI technologies, such as those developed by OpenAI, are unlikely to facilitate breakthroughs on par with groundbreaking theories like those proposed by Nicolaus Copernicus. Wolf's remarks stand in stark contrast to the optimistic views held by prominent leaders in the AI field, including OpenAI’s Sam Altman and Anthropic’s CEO Dario Amodei. When discussing scientific breakthroughs, Wolf refers to transformative ideas that could earn Nobel Prize recognition. He highlights that existing AI chatbots, including ChatGPT, often mirror the biases of their users, responding positively to prompts without challenging them. He elaborates on the inherent limitations of these models, which primarily function by predicting the most probable next word in a sequence. Wolf argues that true scientific innovation stems from a contrarian mindset—scientists question existing paradigms rather than merely predicting likely outcomes. "A scientist's goal is not to forecast the next word; it’s to uncover unexpected truths that defy conventional wisdom," he stated. Wolf's perspective was influenced by an essay from Amodei that claimed AI advancements could significantly accelerate progress in biology and medicine. However, Wolf disagrees, asserting that the current AI capabilities are insufficient for such rapid advancement. He believes these tools might serve more effectively as assistants in the research process, aiding scientists in generating novel ideas rather than making breakthroughs independently. Some initiatives, like Google DeepMind's AlphaFold, are already demonstrating how AI can assist in scientific endeavors by analyzing protein structures, potentially leading to new drug discoveries. Meanwhile, emerging startups such as Lila Sciences and FutureHouse are exploring more ambitious applications of AI to push the boundaries of scientific research even further.