Is AI ushering in a new era of discovery? Labs report advances as skeptics urge caution

Artificial intelligence is edging deeper into the scientific process, with some researchers arguing the field has crossed a threshold from hype to tangible advances. Others, however, insist that while AI can sift data and automate experiments, it has yet to show the kind of creativity that drives paradigm-shifting breakthroughs.

The trajectory stretches back to Adam, a robot that in the 2000s mimicked a biologist by generating questions about yeast, then testing them in a compact automated lab stocked with samples and robotic arms. Adam’s handful of small results are considered the first entirely automated scientific discoveries.

Two decades on, more capable AI systems are working alongside scientists in labs and universities worldwide, and the 2024 Nobel prizes in chemistry and physics went to people who pioneered AI tools. That recognition has not ended the debate. “If you would have asked me maybe a year ago, I would have said there’s a lot of hype,” says Sebastian Musslick, a computational neuroscientist at Osnabrück University in Germany.

Now, he says, “there are actually real discoveries.” Mathematicians, computer scientists and other researchers report breakthroughs using AI agents—systems that can decompose tasks into steps, pull information from the web and generate detailed responses.

Drug companies are building platforms that combine such agents with other AI models to identify new medicines, while engineers are using similar approaches to hunt for materials for batteries, carbon capture and quantum computing. Skeptics see limits that remain stubbornly in place.

Gary Marcus, a cognitive scientist at New York University, argues that a meaningful change in how science is done “is not really happening yet,” calling much of the rhetoric marketing. Today’s systems, he says, excel at finding answers inside boundaries defined by humans—often enormous troves of existing data—but major advances typically require leaps outside those confines.

Large language models that power chatbots and agents draw on vast corpora, including research papers across languages, and can surface obscure connections. But researchers note that discoveries on the scale of continental drift or special relativity demanded imagination and conceptual leaps that current AI does not match.

For now, the technology is altering how people navigate known problems more than it is redefining what questions get asked. Even so, some scientists say they’ve glimpsed what an AI-assisted future might look like. Alex Lupsasca, a theoretical physicist at Vanderbilt University who studies black holes, had identified new symmetries in equations describing the shape of a black hole’s event horizon.

A few months later, in the summer of 2025, he met Mark Chen, the chief research officer for OpenAI, who encouraged him to try a ChatGPT agent running on the then-new GPT-5 pro model. Initially, the system failed to find the same symmetries. After Lupsasca posed an easier warm-up question and then asked again, the agent produced the result he had discovered.

“I was like, oh my God, this is insane,” he says. What comes next may depend on how effectively researchers integrate AI into the full arc of scientific work—from framing questions to designing experiments and interpreting results—without overstating what the systems can do.

For now, the promise is real but provisional: powerful tools that can accelerate parts of discovery, with the creative spark still largely in human hands.