The Constitutionality of Using Computer Models to Determine Criminal Sentences
At a summit in early 2017, an attendee asked Supreme Court of the United States’ Chief Justice John Roberts, “Can you foresee a day when smart machines, driven with artificial intelligences, will assist with courtroom fact-finding, or more controversially even, judicial decision-making?” Chief Justice Roberts responded, “It’s a day that’s here and it’s putting a significant strain on how the judiciary goes about doing things.”
Several technological tools exist that purport to forecast the future. These technologies have also been applied to criminal convicts to determine the likelihood that they will offend again. Such tools are based on algorithms, which are sets of guidelines that tell computers how to model and perform tasks. Algorithms are pervasive, and are used in widely disparate industries, including vehicle global positioning systems, insurance rate setters and credit score evaluators.
One example of software that tries to predict future criminal behavior is COMPAS, which stands for Correctional Offender Management Profiling for Alternative Sanctions. Created in 1998, COMPAS’s algorithms are trade secrets, but we know that it uses six factors from a 137-item questionnaire to assess risk.
The questions represent a range of queries about a defendant: criminal history, gender, age, social life and habits, job history and even parents’ substance use. There are also questions such as “Do you feel discouraged at times?” and “How often did you get in fights while at school?” The questionnaire doesn’t ask a defendant to identify her race.
After evaluating these six factors, COMPAS assigns the defendant a score from 1 to 10. The scores represent the purported risk of his reoffending, and a judge can issue a sentence based on that risk assessment. A defendant defined as medium or high risk, with scores of 5-10, is more likely to receive a harsher sentence, while a low-risk defendant, with a score of 1-4, may be treated more leniently.
Northpointe, COMPAS’s developer, asserts that algorithms are helpful because they help “dispense justice in a more efficient and cost-effective way.” Some have criticized COMPAS, however. In 2016, ProPublica, an investigative news outlet, collected COMPAS scores for more than 10,000 people arrested for crimes in Broward County, Florida, and looked at how many were charged with further crimes in the following two years. Though COMPAS doesn’t ask for a defendant’s race, the study’s results demonstrated that African-American defendants were twice as likely to be incorrectly labeled as higher risk than white defendants.
Two researchers further scrutinized the software’s effectiveness in early 2018. Julia Dressel and Hany Farid conducted a study that found that in small groups of randomly chosen people, use of the defendant’s age and past convictions could predict recidivism with 67% accuracy, a rate slightly higher than COMPAS’s 65% accuracy.
The constitutionality of risk assessment tools like COMPAS was debated in the 2016 case Wisconsin v. Loomis. There, police arrested Eric Loomis and charged him as an accomplice in a shooting. During intake, Loomis filled out a COMPAS questionnaire. He received a score that suggested he was at a high risk of committing another crime because of his background and because he was a registered sex offender.
He pled guilty to eluding an officer and no contest to operating a car without the owner’s consent and was sentenced to six years in prison. Loomis challenged his sentence, arguing that the judge’s consideration of his high-risk COMPAS score violated his right to due process - his constitutional right to fair treatment by the judicial system.
The Due Process Clause of the United States Constitution’s Fifth Amendment provides “No person shall be (…) deprived of life, liberty, or property, without due process of law,” and a long line of cases has held that it requires fair procedures in criminal trials. There is no authoritative list of required procedures, but those that demonstrate fairness include:
· An unbiased tribunal;
· The right to know opposing evidence;
· The right to cross-examine adverse witnesses;
· The opportunity to be represented by counsel; and
· Notice of the proposed action and the grounds for it.
Loomis contended that using COMPAS violated due process because he couldn’t assess COMPAS’s evidentiary accuracy, because he didn’t receive an individualized sentence and because it improperly used gender in providing a risk assessment score.
The Wisconsin Supreme Court disagreed and upheld the use of COMPAS in his sentencing. First, even though he wasn’t allowed to examine how the program uses different factors to predict recidivism because COMPAS’s algorithms are proprietary, since the information the algorithm used came from a questionnaire that he completed and from information available in public records, the court concluded that he had an opportunity to review and ensure the information’s accuracy.
Second, his right to an individualized sentence wasn’t violated because the COMPAS score was just one piece of information that the judge used to reach his sentencing decision. Since it wasn’t the sole determinative factor, his rights to procedural due process weren’t adversely affected.
Finally, the court held that Loomis had not met the burden of proving that the trial court relied on gender as a factor when sentencing him because the trial judge didn’t mention gender when explaining his sentencing rationale.
Perhaps the most interesting and provocative question is whether use of a COMPAS score in sentencing violates the Equal Protection Clause of the Fourteenth Amendment, which prohibits a state from denying “any person within its jurisdiction the equal protection of the laws.”
The 1976 Supreme Court case Craig v. Boren dealt with gender and equal protection violations. There, petitioners challenged an Oklahoma law that prohibited the sale of “nonintoxicating” 3.2% alcohol beer to males under the age of 21 and to females under the age of 18. Oklahoma argued that the law was needed because statistics showed that young men were arrested for drunk driving at more than ten times the rate of young women. Despite these statistics, the Court held that that state statute treating men and women differently was an impermissible gender classification.
Since then, courts have held that a gender classification is only permitted if there is an exceedingly persuasive justification for it. The government rarely achieves this because it is a difficult standard to satisfy.
There is an argument to be made that risk assessment calculators’ use of gender is unconstitutional under the Equal Protection Clause because it doesn’t appear to be “exceedingly persuasive” that using gender in the risk assessment score could prevent criminals from committing crimes in the future and increase judicial efficiency. Thus, it could be an equal protection violation.
The second potential Equal Protection argument is based on race. While COMPAS doesn’t ask for a defendant’s race in the questionnaire, there may be questions that could be “proxies” for racial identification. For example, a COMPAS question such as, “Was your father ever arrested?” could be a proxy for race because it tends to elicit higher “yes” rates from African American respondents.
Still, because the questions are facially neutral, they are subject to much more deferential standards than would be clearly race-based questions. Some showing of discriminatory intent would likely be needed for this to be considered a race-based classification. This means that the defendant challenging his sentence would have to prove that COMPAS included this question for the purpose of racial discrimination; an extremely difficult burden to carry.
Algorithms and risk assessment tools using them can improve the efficiency and equity of the legal system, but there are sure to be debates over their constitutionality as they play an increasingly prominent role in the criminal justice system.
 State v. Loomis, 371 Wis. 2d 235, (2016).
 2 U.S. Const. amend. V
 2 U.S. Const. amend. XIV
 Craig v. Boren, 429 U.S. 190, (1976).
 United States v. Virginia, 518 U.S. 515, (1996).
 United States v. Maples, 501 F.2d 985, (1974).
 Washington v. Davis, 426 U.S. 229 (1976).