Reducing Crime with AI

A recently published study from the National Bureau of Economic Research indicates that artificial intelligence (AI) could help judges determine whether defendants should await trial in jail or at home. An algorithm, created by economists and computer scientists based on data from hundreds of thousands of cases in New York City, was tested on more than 100,000 unrelated lawsuits. The algorithm assigns a risk score based on the current case and the defendant’s criminal history. It also considers age, but does not utilize any other demographic data, including race or socioeconomic status.

A comparison of the results showed that the algorithm was better than human judges at predicting whether or not defendants posed a risk of committing a crime if released. A policy simulation indicated that using the algorithm as a type of “judge’s assistant” could reduce the number of crimes committed by defendants awaiting trial by as much as 25 percent without increasing the jail population. Looked at from another perspective, the jail population could be reduced by 40 percent without an increase in the crime rate. In addition, these gains are possible while reducing the disproportionate number of African-Americans and Hispanics being held in jails.

It’s important to note that this study considered a single variable on which to base algorithmic rules for making predictions. In real practice, judges may consider a broader set of variables, including the severity of the crime and racial inequities. While machine learning can be a valuable adjunct to human decision-making, it is only one component of a broader framework that should include strategies for identifying desired outcomes and reducing bias.