As platforms increasingly rely on automated systems to prevent spam, manage disinformation and limit the spread of fake accounts, what happens when the algorithm makes a mistake? 

University of Michigan School of Information assistant professor Ben Green, an expert on algorithmic fairness, says using human reviewers may not always provide a perfect solution when the machine has made an error.  

“Oftentimes the human overseers are on a tight schedule — they have many cases to review,” he says. “A lot of the cases I’ve looked at are instances where the decision is based on some sort of statistical prediction, but people are not very good at making those predictions, so why would they be good at evaluating them?"

RELATED

Read “What to do when the computer says no” in the Financial Times. 

Ben Green is an assistant professor at UMSI and an assistant professor of public policy at the Gerald R Ford School of Public Policy. Learn more about his research and interests by visiting his faculty profile.