Coworking between humans and AI | Yurec Alert!

Artificial intelligence reduces human error in experiments, but human experts outperform AI when it comes to identifying cause-and-effect relationships or working with small datasets.

To leverage the strengths of AI and researchers, ORNL scientists are collaborating with colleagues at National Cheng Kung University in Taiwan and the University of Tennessee, Knoxville to develop a human-AI collaborative recommender to improve experimental performance. developed the system.

During the experiment, the system’s machine learning algorithm It is described in npj calculation materials, to display preliminary observations for human review. Researchers vote on data to instruct the AI ​​to display similar information or change direction. This is similar to streaming services that generate recommended content based on your activity. After initial guidance, the algorithm is refined to reveal relevant data with little human input.

“The basis of this research is fundamentally not the quantity of data, but the quality of the data we are aiming for,” said ORNL’s Arpan Biswas.

Experiments and autonomous workflows were supported by the DOE-funded Nanophase Materials Science Center, and algorithm development was supported by the DOE-funded MLExchange project to expand machine learning development at national laboratories.