Michigan Tests New Prisoner Classification to Curb Violence

Michigan Tests New Prisoner Classification to Curb Violence

On December 1, Michigan will trial a new prisoner classification system, aiming to predict future violence more accurately by 30%, according to the Michigan Department of Corrections (MDOC). This initiative follows a substantial rise in assaults on U.S. prison staff by 77% and among prisoners by 54% between 2019 and 2024.

Introduction of the Pilot Program

The experiment will begin in selected men’s facilities. Prisoner classifications influence factors such as movement, supervision, activity access, and available programs. The state took action in response to increasing prison violence, collaborating with Recidiviz last year to design this system before considering broader implementation.

Non-AI Driven Approach

Recidiviz, a nonprofit tech organization involved in the system’s development, clarified that the classification tool avoids artificial intelligence in its decision-making. They confirmed that Michigan’s system does not employ AI to determine security levels or release decisions. This approach ensures transparency and focuses on evidence-based, real-time data analysis instead of periodic inputs.

Current and New Classification System

Currently, Michigan uses four security categories: Levels I, II, IV, and V. These levels affect a prisoner’s mobility, supervision, and activity opportunities. In general, higher-security prisoners face stricter restrictions, whereas lower-level prisoners can access vocational training and higher education opportunities.

The new system will use ongoing, real-time data analysis to enhance prediction accuracy about future violence by around 30%. However, MDOC hasn’t detailed the calculation method for this improvement.

Focus on Safety and Evaluation

MDOC Director Heidi E. Washington emphasized that matching security levels to current behaviors and needs helps protect both prisoners and staff. She believes the pilot’s evidence-based and historical data approach will improve facility security. The December pilot aims to test the tool’s effectiveness before considering statewide application.

Concerns Over Algorithmic Bias

Using algorithms in the criminal justice system has raised concerns about transparency and bias. A ProPublica study highlighted issues in a Louisiana risk-analysis system, which differs significantly from Michigan’s proposed security classification. However, similar transparency and data input questions remain, such as how predictive accuracy claims were assessed, whether staff discretion is allowed, and what safeguards against bias exist.

These unresolved issues are vital as Michigan decides whether this safety-focused pilot should be adopted permanently in prisoner classification.

Leave a Reply

Your email address will not be published. Required fields are marked *