The Michigan Department of Corrections (MDOC) is set to trial a pioneering prisoner classification system starting December 1. This initiative aims to improve the precision of predicting future violence among inmates by approximately 30 percent. A review of recent figures reveals a significant increase in prison-related assaults, prompting the state to test this new system.
From 2019 to 2024, assaults on U.S. prison staff surged by 77 percent, while inmate-on-inmate assaults climbed by 54 percent. In response, Michigan’s pilot program will launch in select men’s facilities. Prisoner classifications influence various aspects of prison life, including movement, supervision, and program access.
Recidiviz, the nonprofit behind the system’s development, confirmed to Newsweek that artificial intelligence is absent from the classification tool. Recidiviz assures that AI plays no role in determining security levels or inmate release decisions.
Michigan oversees roughly 32,000 incarcerated individuals. MDOC hopes to address increasing prison violence by revising its classification methodology. Partnering with Recidiviz last year, the state plans to evaluate the system’s effectiveness during the trial period.
Operation of the New Classification System
Michigan’s current security categories for the general prison population include Levels I, II, IV, and V. Level III does not exist. These classifications significantly influence inmate movement, supervision, and activity access. Prisoners under higher security face more restrictions, while those in lower security levels benefit from broader opportunities, such as vocational training and higher education.
The new system promises real-time data analysis for more continuous classification reviews. This shift from scheduled input collection may enhance violence prediction among inmates by about 30 percent, though MDOC’s announcement lacks details on the calculation of this improvement.
AI Exclusion in Recidiviz’s System
Despite offering AI-supported corrections technology, Recidiviz asserts that AI is absent in Michigan’s classification system. A spokesperson clarified, “Michigan’s classification system does not use artificial intelligence. Our tools do not employ AI for security level or release decisions.”
Focus on Safety
MDOC Director Heidi E. Washington emphasized the safety benefits of accurately assigning security levels. “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 stated.
Before broad deployment, MDOC will evaluate the pilot’s outcomes. Broader implementation hinges on the findings of the trial phase.
Transparency and Bias Concerns
Algorithmic risk assessments in the criminal justice system have previously sparked concerns over transparency and racial or socioeconomic biases. For instance, a ProPublica investigation critiqued a similar system in Louisiana for such disparities. While Michigan’s approach differs, queries remain unanswered. These include the types of data used, the basis for predicting accuracy gains, potential staff overrides, and safeguards against bias.
The answers to these questions are pivotal as Michigan assesses whether this pilot could become a permanent fixture in prison security classifications.
This story was produced by Newsweek’s editorial team, utilizing Martyn, our AI assistant. For further information, contact Newsweek editors Robert Greenall, James Debens, and Ben Kelly.