Law in the Internet Society

Risk scores: flawed technology

-- By JessieChao - 20 Nov 2020

Correctional Offender Management Profiling for Alternative Sanctions (COMPAS)

COMPAS is a risk assessment tool that can be used in criminal cases in an effort to determine whether the person will commit a future crime. COMPAS assesses not only risk but also various criminogenic needs such as social isolation, stability/residence, substance abuse, and criminal liability. People are ranked as low, medium, or high risk for each category. Various jurisdictions have adopted COMPAS without testing its efficacy. Although supporters of COMPAS have high hopes for what it can do for our justice system; I have doubts, that COMPAS does more good than harm.

The appeal of risk scores

The US incarcerates more people than any other country. If computers could accurately determine who will commit future crimes, justice could be served more equitably. The court could determine who/how long people should be incarcerated and possibly reduce the number of people in prison (Virginia used risk scores to reduce its rate of incarceration as judges used the tool to send half of the defendants to alternatives other than prison). However, the people making the algorithms are flawed, which makes fair risk scores hard to obtain. If the risk score is too harsh people will unfairly receive a tougher sentence. If the risk score is too lenient dangerous people could be free.

Risk assessment scores are discriminatory against racial minorities and largely not accurate in predicting recidivism rates

Risk assessments are commonly used in criminal courtrooms across the United States to decide anything from bond amounts to sentence length in some states. The risk scores differ greatly depending on if you are white or black. For instance, Borden, a black woman, took a child’s play scooter (left unattended) to try it out and was rated an 8, which is high risk to reoffend. Prater, a white male, stole $86 worth of inventory from a store and he was rated a 3, which is low risk to reoffend. Even though Borden was rated high risk, she has yet to commit another crime; while, Prater is serving an 8-year prison sentence for stealing thousands of dollars.

In a study dealing with 7,000 people who were arrested, the COMPAS risk scores were only reliable in twenty percent of the cases in predicting who would commit a violent crime in the future. The algorithm was more reliable at sixty-one percent, when predicting any future crime which included infractions such as misdemeanors.

This study also showed that the formula was twice as likely to misidentify black people as future criminals. White people were mislabeled low risk more often than black people. This disparity cannot be explained away by prior crimes or the type of crimes that the defendants were arrested for, as a statistical test was run which isolated race from criminal history, age, and gender. In spite of this isolation, black people were seventy-seven percent more likely to be deemed higher risk for committing a future violent crime and forty-five percent more likely to be deemed higher risk for committing any future crime. Unfortunately, there have been few independent studies of the various criminal risk assessments and even fewer that take race into account for the study.

Further making me question the accuracy/fairness of the risk assessments is the fact that defendants are often unable to challenge their risk assessments. The risk assessments do not take into account how a person might have bettered their life such as finding religion, attempts to be more involved in their children’s lives, or the fact that the person is now clean. Essentially, the risk assessments do not give the defendant a voice or take into account that a person can change.

The politics surrounding companies that create these tests

Northpointe is the for-profit company that created the algorithm for COMPAS which was found discriminatory and inaccurate as explained above. It is the most widely used assessment tool in the country. Northpointe vehemently defends the accuracy of the test and will not publicly disclose the calculations used to identify the person’s risk score. This lack of transparency is extremely troubling as the studies have shown that the assessment which impacts people’s lives is flawed. I believe that risk assessments should not be used by the criminal justice system, if at all, unless both parties are permitted to see all of the data that goes into them.

An excellent start. As we have already discussed the ways that the next draft might take additional perspectives into account, I think I should get out of the way and let you take it wherever you think it should go.


Webs Webs

r2 - 28 Dec 2020 - 20:47:27 - EbenMoglen
This site is powered by the TWiki collaboration platform.
All material on this collaboration platform is the property of the contributing authors.
All material marked as authored by Eben Moglen is available under the license terms CC-BY-SA version 4.
Syndicate this site RSSATOM