By Shamena Anwar, John Engberg, Isaac M. Opper, Leah Dion

The use of artificial intelligence (AI) methods to aid with decision making in the criminal justice system has widely expanded in recent years with the increased use of risk assessments. Nowhere has this shift been more dramatic than in the widespread adoption of AI-enabled risk assessment tools to aid in pretrial detention decisions. Despite the promise of pretrial risk assessment tools, the ways in which these tools have been implemented has limited potential progress. The vast majority of jurisdictions that have implemented these tools have essentially provided these risk assessment recommendations to judges in an advisory manner and generally cannot require judges to follow the recommendations when making their pretrial release decisions. Studies indicate that judges frequently ignore the recommendations of the risk assessment instrument; as a result, the adoption of these risk assessment tools has not had much impact on reducing the use of monetary bail and pretrial detention. In this report, the authors investigate the factors that are predictive of whether judges follow risk assessment recommendations and identify the impacts to pretrial detention, public safety, and racial disparities when judges follow the recommendations more often.

Key Findings

• There is wide variation in the extent to which judges follow the recommendations of the pretrial risk assessment when release is recommended. Fifteen percent of judges follow the recommendations less than 40 percent of the time; 22 percent follow the recommendations more than 70 percent of the time.

• When judges deviate from a risk assessment recommendation, they often do so because they penalize certain factors—such as criminal history and charge grade—more than the risk assessment does. The area that a judge represents, their prior occupation, and their age are all factors that are predictive of whether a judge will follow the risk assessment recommendation. However, half of the variation in the rates at which judges follow the recommendations is not explained by any factors that the authors observed about the judge.

• Although judges who follow the risk assessment recommendations at a high rate release more than three times the number of defendants than do judges who follow the recommendations at a low rate, the recidivism rate for defendants assigned to judges in the former group is no different from the recidivism rate for those assigned to judges in the latter group.

• Racial disparities in pretrial detention rates decline as the rate at which judges follow the pretrial risk assessment recommendations increases.

• If judges who follow the risk assessment recommendations at a low rate followed the recommendations more often, more individuals could be released, racial disparities would decline, and there would be no impact on public safety.

2024. 32p.