In less time than it took to read this article, Coscientist, an AI-driven system, autonomously learned about Nobel Prize-winning chemical reactions and designed a successful laboratory procedure to make them. The research team that assembled and tested Coscientist included Carnegie Mellon University chemist and chemical engineer Gabe Gomes, Daniil Boiko, and Robert MacKnight, with support and training from the National Science Foundation.
Building Coscientist’s software and silicon-based parts involved using large language models like GPT-4, as well as equipping it with different software modules. The AI system was able to search public information about chemical compounds, find technical manuals and write computer code to carry out experiments and analyze the resulting data.
The team also introduced Coscientist to high-tech robotic chemistry equipment and taught it to perform tasks such as making robotic liquid handler machines dispense colored liquid into small wells and analyzing liquids using a spectrophotometer. Coscientist demonstrated exceptional abilities, making only a slight mistake before correcting its code on its own.
The breakthrough potential of Coscientist lies in the ability to increase the pace and number of scientific discoveries and improve the reliability of experimental results, showing the power and responsibility of AI in the scientific field.