Michigan will launch a pilot test of a new prisoner classification system on December 1, aiming to enhance predictions of potential violence by about 30%, according to the Michigan Department of Corrections (MDOC). This initiative follows a significant uptick in assaults within U.S. prisons, with staff assaults rising by 77% and prisoner-on-prisoner assaults increasing by 54% from 2019 to 2024, state data indicates.
The pilot will be implemented at selected male facilities, impacting prisoner movement, supervision, and access to programs and activities. The technology organization Recidiviz, which helped develop the system, has clarified that artificial intelligence is not part of decision-making regarding prisoner security levels or release.
Michigan oversees approximately 32,000 incarcerated individuals and initiated this classification overhaul due to rising prison violence nationwide. In collaboration with Recidiviz, the state crafted this new system last year, determining its broader application based on pilot outcomes.
How the New Classification System Will Work
Michigan currently utilizes four security categories for the general prison population: Levels I, II, IV, and V, purposely excluding Level III. These classifications significantly influence prisoner mobility, supervision intensity, and activity access. Higher-security categories typically impose stricter limits, while lower-security prisoners enjoy opportunities like vocational education and training.
This new approach intends to enhance classification reviews by incorporating real-time data analysis, shifting away from fixed intervals for input collection. MDOC anticipates a 30% improvement in the accuracy of violence predictions, though the basis for this figure remains unspecified in their announcement.
Recidiviz Asserts System Does Not Use AI
Despite offering AI-supported technology within its broader corrections framework, Recidiviz insists artificial intelligence is not employed in Michigan’s security classification process. A spokesperson assured Newsweek, “Michigan’s classification system does not use artificial intelligence. AI is never used in Recidiviz tools to make decisions about anyone’s security level or release.”
Michigan’s Aim: Enhancing Facility Safety
MDOC Director Heidi E. Washington explained that accurately placing individuals in more suitable security levels based on current behavior promotes safety for both prisoners and staff. “I am optimistic that this pilot, using evidence-based and historical data, has delivered a dynamic, objective tool to enhance our facilities’ security,” Washington stated.
Instead of rolling out the system across the state immediately, MDOC will evaluate its effectiveness through the December pilot, considering full implementation depending on performance.
Concerns Regarding Algorithmic Risk Assessments
Algorithmic assessments in the criminal justice system typically prompt questions about bias and transparency. A ProPublica investigation into Louisiana’s risk-analysis system found racial and socioeconomic biases, though focused on reoffending and parole eligibility, unlike Michigan’s system for security classification.
These concerns urge Michigan to clarify the data inputs of its assessments, confirm predictive accuracy claims, explore classification override options, and establish bias safeguards before finalizing any permanent adoption.
These details will be integral in determining whether this pilot aimed at enhancing prison safety should become a lasting aspect of prisoner management.
Newsweek’s editorial team, supported by Martyn, their AI assistant, contributed to this article. Contact editors Robert Greenall, James Debens, and Ben Kelly for further inquiries.
Leave a Reply