BEIJING, Dec. 27 (Xinhua) -- An artificial intelligence system called CoScientist, developed by an American research team, recently learned a Nobel Prize-level chemical reaction for the first time and successfully designed an experimental protocol to complete the reaction in minutes. This means that AI is expected to help scientists get faster and more scientific results in the future.
The team led by researchers at Carnegie Mellon University published an article in the British journal Nature, saying that the core of Coscientist is the multimodal large language model GPT-4, and its completed "palladium-catalyzed cross-coupling reaction" has been widely used in scientific research, medicine and electronics industry around the world. In 2010, three scientists from the United States and Japan were awarded the Nobel Prize in Chemistry for their outstanding contributions to this reaction-related field.
According to the researchers, Coscientist was able to search the Internet for publicly available information about compounds, find and read technical manuals on how to control robotic lab equipment, write computers** to conduct experiments, and analyze the validity of the results. In the experimental operation, Coscientist can control the high-tech robots and other experimental equipment in the laboratory to accurately suck, spray, heat, shake the small liquid sample, etc., and finally successfully synthesize the target chemical substance.
Before completing the "palladium-catalyzed cross-coupling reaction", the research team tested the different software modules equipped with the coscientist individually. In one of the tests, commonly used substances such as aspirin, acetaminophen and ibuprofen were successfully produced if the steps designed by the coscientist were followed.
For the first time, AI has planned, designed, and executed complex chemical reactions invented by humans, researchers say. There is an iterative process of trying, failing, learning, and improving in scientific research, and AI is expected to greatly accelerate this process, as the latter can "think" around the clock and make up for the shortcomings of human scientists. (ENDS).