Yikes! “AI can predict which criminals may break laws again better than humans”

Science (the journal!) has this heading on its RSS feed page, which makes me wonder if they have been paying any attention to the well-documented issues with AI driven “justice”.

“…some research has given reason to doubt that algorithms are any better at predicting arrests than humans are.”

Among other issues, the study compared volunteers with COMPAS‘ or LSI-R predictive abilities for predicting violent crime behaviour, based on the same covariates. Volunteers, not experts! And the algorithms are only correct 80% of the time which is a terrible perfomance when someone’s time in jail depends on it!

“Since neither humans nor algorithms show amazing accuracy at predicting whether someone will commit a crime two years down the line, “should we be using [those forecasts] as a metric to determine whether somebody goes free?” Farid says. “My argument is no.””