Researchers at Chalmers University of Technology in Sweden have created an AI scientist capable of independently formulating scientific hypotheses, planning experiments and interpreting results. The system operates in a closed loop, using brewer's yeast Saccharomyces cerevisiae as an example.
It was built on extensive data: the yeast genome, its metabolism and previous studies. For a human, the volume of information is too great, but the AI successfully identifies promising questions, proposes experiments, evaluates outcomes and refines its understanding based on new data.
"Our AI scientist does not merely help make decisions, but actively creates new scientific knowledge," notes postdoc Ievgeniia Tiukova from the Chalmers Department of Life Sciences.
The development combines large language models, automated reasoning and laboratory automation. Professor Ross King, the senior author of the study, compares it to self-driving cars: the AI processes information, draws conclusions and acts.
Autonomous laboratories, according to King, will be able to systematically study biological systems far faster than today. They will collaborate with humans, accelerating discoveries in biology, medicine and biotechnology, as well as optimizing the use of resources.
For now, such systems complement scientists, taking on the routine cycles of hypothesis generation and testing. Humans remain key to setting priorities, interpreting significance and exercising ethical oversight.
King previously created the first robot scientists, Adam and Eve. The new work is published in the Journal of the Royal Society Interface. How will the role of humans in the laboratory of the future change when AI takes on more and more routine tasks?
