Chalmers AI scientist generates hypotheses and runs yeast experiments in a closed loop
Researchers in Sweden combined language models, automated reasoning and lab robotics to let an AI propose, test and interpret experiments on brewer's yeast.
Researchers at Chalmers University of Technology in Sweden have built an AI scientist that can generate scientific hypotheses, design experiments and interpret the results, with minimal human intervention. The university announced the work on September 30.
How it works
The team combined large language models, automated reasoning and laboratory automation into a closed-loop AI laboratory. It works on brewer's yeast, Saccharomyces cerevisiae. The AI was given scientific knowledge to start from, including the yeast's genome, metabolism and earlier studies. From there it identified promising biological questions, recommended experiments to test them, evaluated the outcomes and refined its understanding based on new evidence.
Postdoctoral researcher Ievgeniia Tiukova says it is too much information for a human to analyze, and that the AI scientist "actively generates new scientific knowledge" instead of only supporting decisions. She compares the development to self-driving cars, where AI processes information, draws conclusions and acts. The robot scientist Eve, designed for drug discovery, was updated with large language models and automated reasoning for this work.
What the authors say about the future
Professor Ross King, the senior author, says autonomous laboratories could investigate biological systems much faster than is possible today and help across biology, medicine and biotechnology. Both authors stress that for now the AI will augment scientists, not replace them. Humans stay essential for setting research priorities, judging broader significance and ensuring ethical oversight.
The study is titled "Agentic AI integrated with scientific knowledge: laboratory validation in systems biology" and appeared in the Journal of the Royal Society Interface. Authors come from Chalmers, the University of Gothenburg and the University of Cambridge, and funding includes WASP, the UK's EPSRC and the Swedish Research Council Formas.
Why it matters
A lot of AI talk is about chat and coding. Here an agent closes the loop with the physical world: it thinks, then a robot runs the experiment, then the AI reads the result. Self-driving labs could shorten the time from question to answer, and use lab resources better.
Dany's take
This is the kind of agent story I would like to see more of. The tasks are well defined, the results can be checked in a lab, and humans keep the priorities. I would still read the paper before calling anything a discovery, because the press release does not list the specific findings. Closed-loop science is a trend to watch, and yeast is a very sensible place to start.
Source: chalmers.se