Eric Loomis COMPAS prison sentencing

The sentencing of Eric Loomis for driving a car used in a shooting by a judge partly relying on the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) tool led to Loomis lodging an appeal on the basis that the tool was a black box, the score assigned to him could not be assessed, and that due process had not been followed.

Loomis v Wisconsin

Per George Washington University's ETI AI Litigation Database, COMPAS risk assessments are based on data gathered from a defendant's criminal file and from an interview with the defendant, and predict the risk of pretrial recidivism, general recidivism, and violent recidivism.

A registered sex offender, Loomis scored high on all three risk measures, and was sentenced to six years in prison. The judge said he had arrived at his sentencing decision in part because of Mr. Loomis’s rating on the Compas assessment. Loomis then filed a motion requesting a new sentencing hearing, which was denied by all three levels of Wisconsin courts.

Nonetheless, the court advised that judges presented with COMPAS risk scores understand that: 

1. Tthere is no disclosure of how factors are weighed or risk scores are determined; 

2. That the assessment compares defendants to a national sample and no cross-validation study of Wisconsin defendants has been completed; 

3. Some studies have raised questions about whether COMPAS disproportionately gives minorities higher risk scores; and 

4. Risk assessment tools must be updated due to changing populations.

The judge also ruled that the use of COMPAS does not deny due process as the court ultimately imposes sentences on the basis of all of its knowledge of the defendant, including their criminal histories.

Algorithmic justice

The incident raised broader questions about the opacity, efficacy, and fairness of the tool and of tools similar to it, such as the Ontario Domestic Assault Risk Assessment (ODARA) and the Virginia Non-violent Risk Assessment (NVRA).

It also prompted discussions about the use of big data and algorithms to predict crime (aka 'predictive policing').