Discussion

Michigan prisons plan AI-assisted security classifications, raising questions about the evidence

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#10498

Michigan plans to pilot AI-assisted decisions about where incarcerated people are housed, according to Futurism. The system could influence security classifications and access to programmes, making the quality of its data and the limits of human oversight matter well beyond the software itself.

Watch Desk analysis

What happened

Futurism reports that the Michigan Department of Corrections plans to begin a pilot with Recidiviz software in several men’s facilities in December. The state currently uses five housing levels, with screenings conducted by staff appointed by each facility’s warden, the report says.

The software is described as helping optimise placement and identify people for reclassification, release or state programmes. Futurism says the department expects the tool to improve its ability to predict future violence by about 30 per cent, but the report says it is unclear how that figure was calculated. That is a consequential claim, not a scorecard readers can yet inspect.

Why it matters

Housing classifications can affect which programmes and resources people can access, as well as who they are housed alongside, according to Futurism. A model used in that process could therefore shape daily conditions and opportunities, even if it does not make the final decision on its own.

Futurism points to a ProPublica investigation of a risk-assessment programme in Louisiana that found racial and socioeconomic bias in the data used to classify incarcerated people. That is a warning about a comparable system, not evidence that Michigan’s tool has the same flaws. The crucial questions here are what data Michigan will use, how its predictions will be tested, and how people can challenge a classification.

Our read

This is a meaningful AI deployment in a setting where a bad classification can have practical consequences for people who have little power to contest it. The reported 30 per cent improvement deserves scrutiny before it earns anyone’s trust. An impressive percentage without a clear baseline or method is still a number wearing a rather important hat.

What to watch

  • What data the Michigan pilot will use, and whether the state will publish its evaluation method.
  • Whether staff can override recommendations and how incarcerated people can challenge decisions.
  • What the pilot measures, including errors and differences in outcomes across groups.

Discussion spark: Should Michigan use AI to inform prison housing decisions before it publishes the data, evaluation method and appeal process, or should those safeguards come first?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.