Criminal court systems across the country are using Northpointe, a Minority Report-esque algorithm designed to predict an offender’s likelihood to commit another crime in the future, to aid decisions about things like eligibility for bail and jail-diversion programs. Unfortunately, just like many non-algorithmic elements of the justice system, Northpointe seems to have a thing against black people.
ProPublica published an investigation into Northpoint’s effectiveness in predicting recidivism, as evidence of racial bias in this effectiveness, and found that, after controlling for variables such as gender and criminal history, black people were 77 percent more likely to be predicted to commit a future violent crime and 45 percent more likely to be predicted to commit a crime of any kind. The study, which looked at 7,000 so-called risk scores issued in Florida’s Broward County, also found that Northpointe isn’t a particularly effective predictor in general, regardless of race: Only 20 percent of the people it predicted to commit a violent crime in the future ended up doing so.
Many other jurisdictions use Northpointe or other systems like it. In every New York county outside of New York City, Northpointe is used to assess inmates’ probation eligibility, and in Wisconsin, it is used during “each step in the prison system,” including sentencing. A recent federal sentencing reform bill would require the use of risk assessment in the federal corrections system.
According to ProPublica, Northpointe’s calculations are based on a list of 137 questions, such as “Was one of your parents ever sent to jail or prison?” and “How often did you get in fights while at school?” The questionnaire does not directly address race, but it contains sections that might correlate with race, such as questions about poverty. The founder of the company told ProPublica that it is difficult to construct an accurate questionnaire without including such factors, and a company spokesperson said it disagreed with the finding that Northpointe’s scores were biased.
The bias is almost certainly unintentional, but the proof is in the pudding. Included in the investigation are side-by-side comparisons of white offenders versus black ones who were convicted of similar crimes. In one case, a white man arrested for drug possession who had one prior arrest was rated a risk level 3, a relatively low score, and went on to be arrested three more times. A black man who was also arrested for drug possession and had one prior arrest was rated a 10. Since then, his record has been totally clean.