Predictive Policing Is Terrible At Both Predicting And Policing
Minority Report was based on the idea that with enough data (and some freaky psychics), you can stop crimes before they even happen. But surely that’s too crazy to try in real life, right? Yes, absolutely, but when has that ever stopped humanity? Predictive policing programs have popped up all over the United States, and one of the most popular ones is COMPAS, which provides courts with “risk scores.” COMPAS looks at a person’s criminal history and life circumstances, and comes up with a score that reflects how likely that person is to reoffend, like a terrible real-life RPG stat. Courts use these scores to determine bail, and sometimes sentences.
Now, we’re huge nerds, so of course we love statistics. The problem is that so far, predictive policing only interprets those stats like someone who doesn’t know what numbers are, but is pretty sure they hate them. A ProPublica expose on COMPAS revealed that African-Americans were “45% more likely to be assigned higher risk scores than violent offenders.” In one of their examples, an 18-year-old black kid who stole a scooter was assessed as a high risk to reoffend, while an adult white repeat offender with multiple armed robbery convictions was considered a low risk. The white guy ended up going back to jail, while the “high risk” kid hasn’t reoffended.