AI and machine learning are being rapidly adopted in prisons and jails across the US, with some officials hoping that AI-powered machines, like Tesla's Optimus robot, could someday address staffing shortages. However, critics are concerned about opaque data collection, privacy violations, and inherent biases in these technologies, which are often used for far more critical tasks than serving snacks.
Risk assessment tools, some of which have been in use for over a decade, are being enhanced with AI to predict violence and recidivism. These "AI violence predictors" analyze vast datasets, including an incarcerated person's age and past violent incidents. For example, California uses COMPAS to assign recidivism scores to inmates, a practice that civil rights groups worry could perpetuate bias. Some correctional facilities, such as in Maricopa County, Arizona, and in Colorado and Alabama, are also exploring "biometric behavioral profiling" combined with AI to prevent in-custody deaths and medical emergencies.
Louisiana has implemented a controversial AI system called TIGER (Targeted Interventions to Greater Enhance Re-entry) which automatically blocks approximately 13,000 prisoners, nearly half of its incarcerated population, from parole hearings if they receive a "moderate risk" score or higher. This led to a 78% drop in parole approvals, from 32 per month in 2023 to six per month currently. The TIGER algorithm weighs factors like decades-old parole violations, effectively negating current rehabilitation efforts. Meanwhile, California identified six "high-risk automated decision systems" in use for criminal justice, including those for evaluating recidivism and screening unemployment claims for fraud.
Despite concerns, some AI applications show promise in reducing recidivism and associated costs. Diversion programs that use improved decision-making tools, including machine-learning to identify individuals most likely to succeed, have been effective. The Adult Redeploy Illinois program, for instance, has reduced recidivism rates by 20% or more in some cases, and its interventions cost an average of $5,000 per person annually, compared to $49,000 for incarceration, reportedly saving an estimated $83 million in total avoided costs for fiscal year 2025.
While AI's potential to improve efficiency and reduce recidivism is recognized, the implementation of these high-stakes systems by states like Louisiana and California has sparked alarm among civil rights advocates. They argue that the technology can exacerbate existing biases, violate attorney-client privilege (as seen with Securus lawsuits over recorded calls), and make life-altering decisions based on algorithms that are difficult to challenge or understand.
