Policy

Michigan Prison Classification System Sparks AI Definition Debate

Corrections officials and their contractor say new predictive software doesn't use artificial intelligence, but experts disagree on where to draw the line.

Omega Editorial · November 18, 2026 · 3 min read

Key takeaways

  • Michigan's new prison classification system claims 30% better violence prediction but officials deny it uses artificial intelligence.
  • Northwestern professor Sylvester Johnson says the automated decision system clearly qualifies as AI regardless of official terminology.
  • The $4 million Recidiviz system analyzes ten years of data to weight factors like assault history and education in security classifications.
  • Experts warn that proprietary algorithms may reduce transparency and make it harder for inmates to challenge classification decisions.
  • Prison systems nationwide are rapidly adopting predictive technologies to address chronic staffing shortages despite ongoing controversy.

Automated predictions without AI?

Michigan's Department of Corrections is launching a new prisoner classification system in December that officials say will improve predictions of future inmate violence by roughly 30 percent. Yet both the agency and its contractor, Utah-based Recidiviz, insist the technology does not involve artificial intelligence — a claim that has drawn skepticism from multiple experts in the field.

The $4 million system replaces manual scoring forms with automated calculations that update in real time based on factors like escape attempts, assault history, life sentences, and educational achievements. Staff receive alerts when an inmate's classification needs review, though humans make final decisions on security levels ranging from 1 (lowest) to 5 (highest).

Recidiviz developed the new scoring weights by analyzing a decade of Michigan classification data, determining which factors actually correlate with violent behavior and how much each should count. The company describes this as traditional data analysis using correlational and regression statistics, not machine learning or generative AI.

Why it matters

Classification decisions directly affect inmates' daily lives — determining housing arrangements, movement privileges, visitation rights, and program eligibility. As cash-strapped prison systems nationwide adopt predictive technologies to address staffing shortages, the debate over what constitutes AI has real consequences for transparency, accountability, and the ability to challenge potentially biased algorithmic decisions.

Experts challenge the distinction

Sylvester Johnson, a Northwestern University professor specializing in artificial intelligence and society, said the Michigan system clearly qualifies as AI despite official denials. He defines AI in its basic form as "machine systems that can take inputs, and based on those inputs, and based on a set of instructions that might be fixed commands, they can produce an output."

Johnson, who served as the 2024 Kluge Chair in Technology and Society at the Library of Congress, called it "definitely an automated decision system" and a form of AI, "even if they don't want to use the term."

Siwei Lyu, director of the Institute for Artificial Intelligence and Data Science at the University at Buffalo, took a more cautious position. He noted that AI is "a broad and sometimes loosely used term," and that the classification module itself appears to use traditional points-based algorithms. However, he acknowledged that developing the scoring rules through statistical analysis of historical data could involve machine learning techniques.

Hayden Smith, a University of South Carolina criminology professor who studies prison technology, said it appeared Recidiviz used some form of AI to interpret the decade of data, though he would need more detailed methodology to make a definitive assessment.

Transparency and bias concerns

Johnson raised concerns about potential biases embedded in the algorithm and reduced transparency compared to the old standardized forms. Because the software is proprietary, inmates may find it harder to understand and challenge their classification scores.

Recidiviz spokesperson Maddy Woodle countered that the new system is actually simpler to understand than the complex old one. She said data points include age but not race, gender, or geographic origin, and that point values for each factor will be clear to both staff and prisoners.

The technology adoption reflects broader trends in corrections. Smith noted that prison systems accelerated their embrace of technology after the COVID-19 pandemic, with vendors competing for market share in body cameras, drones, and predictive systems.

Yet using technology to forecast human behavior remains controversial, Smith acknowledged, inevitably evoking comparisons to the 2002 film "Minority Report" and its fictional "Precrime" unit.

These details were first reported by Paul Egan of the Detroit Free Press.

  • #prison technology
  • #artificial intelligence
  • #predictive algorithms
  • #criminal justice
  • #recidiviz
  • #michigan corrections

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Policy