Incarcerated men in Michigan are involved in employment counseling, where their access to job training and education depends on their security classification. The state is now preparing to test new technology that could influence security classifications and potentially improve predictions of future prison violence by about 30%.
The system, developed by Recidiviz, is not based on artificial intelligence. Instead, it uses predetermined rules, and the final decisions are made by corrections staff. This approach seeks to refine the assessment of each individual’s behavior and needs, providing a more accurate security classification.
Current Security Levels
The Michigan Department of Corrections currently oversees approximately 32,000 incarcerated individuals. Security levels, ranging from I to V, influence not only living arrangements but also access to educational programs. For inmates, a change in security level can significantly impact daily life.
The state uses a manual scoring process to determine security levels, which affects not only housing but also access to vocational and educational opportunities. Michigan aims to improve this process with ongoing reviews using updated records rather than at fixed intervals.
Upcoming Changes with New System
The new system will be tested in men’s prisons starting December 1. It will enable continuous security reviews based on recent prison records. The software factors in current disciplinary incidents and any completed rehabilitation programs to determine if an inmate qualifies for a less restrictive classification.
Heidi E. Washington, Director of the Michigan Department of Corrections, supports these changes, believing they could improve safety through better classification fitting current behaviors. This system intends to identify changes in behavior more promptly, potentially leading to safer conditions both for staff and inmates.
Clarification on AI Usage
Recidiviz clarified that AI is not used in Michigan’s new classification system. The classification relies on a points-based formula, which staff can review and adjust when necessary. This ensures that decisions are grounded in transparent procedures.
Despite the growing presence of AI across industries, this system employs a more traditional software approach. While AI helps manage case notes and organizing meetings within Recidiviz’s broader technology platform, it does not influence security levels or releases.
Concerns and Challenges
The expected 30% improvement in predicting future violence drew attention, but clear data supporting this figure remains unexplained. Accuracy issues may arise if the underlying information is flawed, affecting the outcomes and decisions based on these scores.
Michigan faces logistical challenges beyond software advancements. Union representative Byron Osborn highlighted a shortage of high-security beds, noting that even if classifications improve, effective housing solutions remain crucial to addressing safety concerns.
Plans to add 870 high-security beds are underway, but completion will take time. The state needs to address these infrastructure issues to realize the benefits of improved security classifications fully.
As Michigan tests this system, the ongoing evaluation will determine its effectiveness in enhancing prison safety and fairness in inmate classifications.
