AI

Michigan deploys Recidiviz AI platform to classify and sort prisoners

Michigan Department of Corrections launches the Recidiviz AI algorithm to assess inmate risk levels, replacing special review boards.

·2 min read
Michigan deploys Recidiviz AI platform to classify and sort prisoners

The Michigan Department of Corrections is launching a pilot program using software from the non-profit organization Recidiviz in several men's prisons. Originally created as a Google project to standardize criminal justice data, the system will classify individuals into five custody levels to determine their environment and rehabilitation programs.

  • Recidiviz AI software rollout begins in Michigan prisons starting in December.
  • The system classifies inmates into five security and rehabilitation tiers.
  • Officials claim a 30% boost in violence prediction accuracy, details withheld.
  • Critics warn of historical data bias and potential discrimination.

How the new algorithm replaces traditional committees

Previously, decisions regarding prisoner housing and placement were handled by a committee of staff appointed by the prison warden. Now, the AI algorithm will serve as a key factor in optimizing prisoner placement. The software determines the security level and environment an inmate will experience throughout their entire sentence.

Department representatives claim the AI will improve the prediction of violence risk by approximately 30 percent. However, officials have declined to disclose the specific data sources and metrics used to arrive at this performance evaluation, raising questions about transparency.

Read alsoGoogle and OpenAI agree on a unified robots.txt standard for AI

Algorithmic bias concerns and past precedents

The issue of algorithmic bias in correctional systems already has notable precedents. An investigation by ProPublica in Louisiana revealed that a similar program classified a 70-year-old blind wheelchair user with a flawless disciplinary record as a "medium risk" of recidivism, automatically blocking his parole.

Although the Michigan system is designed for housing placement rather than parole decisions, experts fear underlying issues. Using historical data laden with socioeconomic and racial disparities risks exacerbating the existing overcrowding and injustice crisis within the prison system.

Developer goals and unanswered transparency questions

Recidiviz describes its core mission as providing tools to identify individuals eligible for educational programs, job retraining, or release, while coordinating access to resources like temporary housing and employment. The platform aims to streamline rehabilitation efforts for inmates.

However, the lack of independent validation and the authorities' refusal to share data collection specifics leave critical questions unanswered. It remains to be seen whether this algorithm will genuinely aid rehabilitation or simply become another mechanism of algorithmic discrimination.

Source: iXBT.com

Related articles