Software used across the US to help predict whether people who end up in courtrooms, jails, and prisons are likely to commit more crimes appears to be biased against black people, according to a sweeping, thorough ProPublica investigation.
The software, which is designed by several for-profit and nonprofit groups, looks at a series of risk factors in people’s lives and assigns them “risk scores” as to how likely it thinks they are to reoffend. Those scores get used in jails, prisons, and courtrooms.
In Arizona, Colorado, Delaware, Kentucky, Louisiana, Oklahoma, Virginia, Washington and Wisconsin the scores are introduced as part of the sentencing process.
ProPublica studied data from 7,000 people arrested in Broward County, Florida and found three key things:
The scores are not very good at predicting who will reoffend.
White people were much more likely to get low scores and then reoffend.
Black people were much more likely to get high scores and not reoffend.
It seems likely that this is a result of the software taking into account factors like wealth and social marginalization that correlate heavily with race. It’s not the first example we’ve seen lately of an calculation not designed explicitly for racist purposes producing racist results, though it is the one we’ve seen with the highest stakes.
Northpointe, the company that evaluated risks in Broward County, disputes ProPublica’s analysis, saying these claims do not “accurately reflect the outcomes from the application of the model.” That said, the company would not disclose how it arrived at its risk scores, so it’s tough to know what’s driving this algorithm.
ProPublica’s report is full of jarring comparisons between the way the algorithm rates black people and white people. You can read the full report here.
Northpointe did not immediately respond to a request for comment.
Read the original article here.