Chemistry’s Valentin Rodionov finds AI models often rely on flawed or discredited scientific research

AI agents are not good at interpreting science. What can be done?

Chemical & Engineering News: Valentin Rodionov, professor of chemistry in the College of Arts and Sciences, found that 30 AI models recommended building on flawed or discredited research about 93% of the time. Researchers are proposing solutions, including credibility indexes, metadata instructions and uncertainty labels, to help AI distinguish reliable science from retracted or questionable studies.