AI could help make smarter inmate parole evaluations at U.S. prisons, study says

SACRAMENTO, Calif. — Whether or not an inmate is truly reformed may one day be partially determined by artificial intelligence, new research suggests.

The United States has long been a leader when it comes to overcrowded prisons, currently ranking sixth in that category on a global scale. So, over the past decade, there has been an effort to cut down on the nation’s high incarceration rates – without impacting public safety, of course.

Consequently, many parole boards have started making risk-based parole decisions, or releasing inmates deemed low-risk of committing a crime after release. This new study’s authors, a team of researchers from the UC Davis Violence Prevention Research Program and the University of Missouri, Kansas City, set out to gauge the effectiveness of the current risk-based parole system. This was accomplished using a machine learning program that analyzed parole data from New York.

Ultimately, the AI program concluded the New York State Parole Board could safely grant parole to more inmates.

“We conservatively estimate the board could have more than doubled the release rate without increasing the total or violent felony arrest rate. And they could have achieved these gains while simultaneously eliminating racial disparities in release rates,” says lead study author Hannah S. Laqueur, an assistant professor in the Department of Emergency Medicine, in a university release.

There are 1.2 million people in U.S. prisons

According to the Bureau of Justice Statistics, at the end of 2021, the total U.S. prison population for state, federal, and military correctional facilities was 1,204,300.

The research team utilized the machine-learning algorithm SuperLearner to predict potential arrests, including violent felony arrests, within three years of an inmate being granted parole. That AI program considered a total of 91 variables to predict individual crime risk. Examples of such variables include age, minimum and maximum sentence, prison type, race, time in prison, and any previous arrests.

From there, risk-prediction models were modeled using data pertaining to 4,168 individuals who had been released on parole in New York between 2012 and 2015.

Several tests were put together aimed at validating the algorithm across the full population of individuals up for parole. This population included inmates who had hearings and were denied parole by the board but were eventually released at the end of their maximum sentence (6,784 individuals).

The AI program noted that the predicted risks for those denied parole and those released were very similar. This suggests low-risk individuals may have remained incarcerated, while high-risk individuals ended up being released.

Study authors stress that they are by no means advocating for machines and AI to totally replace human decision makers in these matters. Instead, they believe AI can help identify problems in current parole systems.

“This study demonstrates the utility of algorithms for evaluating criminal justice decision-making. Our analyses suggest that many individuals are being denied parole and incarcerated past their minimum sentence despite being a low risk to public safety. We hope that by providing data on predicted risks, we can aid reform efforts,” Prof. Laqueur concludes.

The study is published in the Journal of Quantitative Criminology.

YouTube video