Artificial intelligence has been taught to conduct laboratory experiments — the speed is 10,000 experiments per day

Scientists have created an artificial intelligence-based system that allows robots to conduct up to 10,000 scientific experiments a day without human intervention. The system, called BacterAI, can significantly accelerate discoveries in various fields of science, such as medicine, agriculture, and ecology.

In a recent study published in the journal Nature Microbiology, the team successfully used BacterAI to study the metabolic processes of two microorganisms associated with oral health.

A team of researchers from the University of Michigan, led by Professor Paul Jensen, was trying to determine which amino acids are necessary for the growth of beneficial bacteria in the mouth. However, this was difficult to do due to the large number of possible combinations of the 20 available amino acids.

To solve this problem, the BacterAI system was developed, which tested hundreds of amino acid combinations every day and adjusted them based on the results of previous experiments. In nine days, BacterAI learned to accurately predict the necessary combinations of amino acids for bacterial growth in 90% of cases.

Automated experimentation with BacterAI can be used not only in microbiology but also in other fields of science, where researchers can formulate questions as puzzles for AI to solve by trial and error.

BacterAI is a unique achievement in the field of AI because it is able to create its own dataset through a series of experiments, unlike the traditional approach where ready-made data is loaded into a machine learning model.

The system uses the results of previous tests to make predictions about which experiments can provide the most information. In less than 4,000 experiments, BacterAI has figured out most of the rules for proper «feeding» of bacteria.

The development of BacterAI led by Professor Paul Jensen is a big step forward in the use of AI to accelerate various scientific discoveries. As the research has shown, BacterAI can provide important information about the microbial world, in particular, about those bacteria that affect our health.