Swedish researchers develop AI system that can run full scientific discovery cycle

Researchers in Sweden have developed an artificial intelligence system designed to carry out much of the scientific discovery process with limited human intervention, from generating hypotheses and designing experiments to analysing results and refining future research.

The system combines several large language models, scientific databases and laboratory robotics in a closed-loop process that allows it to repeatedly generate and test ideas. The research was conducted by scientists at Chalmers University of Technology and the University of Gothenburg.

In a study published in the Royal Society journal Interface Focus, the researchers tested the system using a database containing about 60,000 phenotypical, physiological and metabolic relationships involving Saccharomyces cerevisiae, a yeast commonly known as brewer’s or baker’s yeast.

The AI system analysed the available information and generated almost 2,000 testable predictions about how nutrients could affect yeast growth and resistance to stress.

It then assessed the proposed hypotheses, selected appropriate comparison controls and converted the experimental plans into machine-readable instructions that could be followed by laboratory equipment.

After the experiments were completed, the system analysed the results and assessed whether its predictions had been supported. It could then update, refine or reject hypotheses and use the new evidence in subsequent research cycles.

“It is too much information for a human to analyse, but our AI scientist could identify promising biological questions, recommend experiments to test them, evaluate experimental outcomes and iteratively refine its understanding based on new evidence,” said Ievgeniia Tiukova, a researcher at Chalmers University of Technology and one of the study’s authors.

Tiukova said the system went beyond acting as a tool to assist researchers by generating new scientific knowledge through its experimental process.

The researchers said increasing automation could change how scientific research is conducted as artificial intelligence, robotics and high-throughput technologies continue to develop.

Ross King, senior author of the study at the University of Gothenburg, said AI scientists could work alongside human researchers to accelerate discoveries in biology, medicine and biotechnology.

The researchers said automated systems could also reduce some sources of variation in experiments, including human error, incomplete recording of procedures and differences in laboratory conditions. They added that automation could allow scientists to investigate more research questions while making more efficient use of laboratory resources.

However, the researchers stressed that autonomous systems would not remove the need for human scientists. People would remain responsible for setting research priorities, considering wider scientific and social contexts, developing research programmes and ensuring that experiments comply with ethical requirements.

King said future autonomous laboratories could become increasingly capable of working alongside researchers and contribute to efforts to address complex questions in biology and medicine.

The development comes as international organisations consider how artificial intelligence should be governed in scientific and health research.

The World Health Organization has issued recommendations for researchers, ethics committees, regulators, funders and policymakers on the responsible use of AI in health research, stressing the need for safeguards covering human dignity, rights and equity.