10
Loading video...
By Megan Cartwright
Associate News Editor
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
Michigan will begin testing a new technology-assisted prisoner classification system on December 1 that officials say could improve predictions of future violence by about 30 percent, the Michigan Department of Corrections (MDOC) announced.
The pilot arrives after assaults on U.S. prison staff rose 77 percent and assaults between incarcerated people increased 54 percent from 2019 to 2024, according to figures cited by the state.
The trial will begin at selected men’s facilities, where classifications can affect prisoners’ movement, supervision and access to activities and some programs.
Read More on AI
Newsweek reached out to the MDOC via email for comment on Monday.
Michigan supervises about 32,000 incarcerated people and says it began overhauling its classification approach in response to rising violence in prisons nationwide. The state brought in nonprofit technology organization Recidiviz last year to help develop a more dynamic system before deciding whether it should be used more widely.
How the New Classification System Will Work
Michigan currently uses four general-population security categories: Levels I, II, IV and V. There is no Level III.
Those classifications have real consequences. They help determine how freely someone can move, how closely they are supervised and what activities they can access. Higher-security prisoners generally have tighter restrictions, while people at lower levels can have access to a broader range of opportunities such as vocational training and higher education.
...
Under the new approach, officials say classification reviews will incorporate real-time data that can be analyzed on an ongoing basis rather than relying on inputs collected at scheduled intervals.
MDOC says that change is expected to make its predictions of future violence approximately 30 percent more accurate. Its public announcement, however, does not explain how that figure was calculated.
Where AI Fits In
The technology comes from Recidiviz, a nonprofit organization whose corrections platform uses data, automation and AI. According to the company's website, its wider suite of tools can help agencies identify people eligible for programs, reclassification or release, as well as connect individuals with services including education, transitional housing and employment.
However, there is an important distinction. Michigan's announcement describes its new classification system as using "evidence-based and historical data," but does not itself characterize the risk-classification tool as artificial intelligence. Recidiviz describes its broader corrections platform as using AI.
That leaves an important question over exactly what role AI plays in Michigan’s individual risk assessments and how much weight the technology will carry in determining security classifications.
Newsweek contacted Recidiviz via email for comment on Monday.
Michigan Says the Goal Is Safety
MDOC Director Heidi E. Washington said placing people at security levels suited to their current behavior and needs protects incarcerated people and corrections staff.
"I am optimistic that this pilot’s use of evidence-based and historical data has created a tool that is dynamic, objective and will strengthen the security of our facilities," Washington said.
Rather than immediately introducing the system throughout Michigan, MDOC says it will use the December pilot to establish how well the tool performs before considering wider adoption.
Questions Remain Over Algorithmic Risk Assessments
Using algorithms to assess people in the criminal justice system has previously raised questions about transparency and bias.
A ProPublica investigation examined a different risk-analysis system used in Louisiana and raised concerns about racial and socioeconomic disparities in the data used to assess people. That system focused on reoffending and parole eligibility, making it materially different from Michigan’s proposed approach to prison security classifications.
The comparison nevertheless highlights questions Michigan has yet to answer publicly in its announcement, including what information will feed into its assessments, how the claimed improvement in predictive accuracy was established, whether staff can override classifications and what safeguards will be used to detect bias.
Those details could become crucial as Michigan decides whether a pilot intended to make prisons safer should become a permanent part of how incarcerated people are classified.
Newsweek’s reporters and editors used Martyn, our AI assistant, to produce this story. Learn more about Martyn here. Contact Newsweek editors for this story: Robert Greenall and Ben Kelly
View 0 Comments
Submit Correction
View Editorial & AI Guidelines