Rethinking AI Implementation in Criminal Justice

Abstract Artificial intelligence’s integration into U.S. criminal justice systems is now a given rather than a possibility. The question now expands from whether AI should be used to what the appropriate role for AI is in criminal justice and the guardrails that need to be established, especially in these still early stages of development. More specifically, is it proper for AI technology to be tested in the first instance on the general public or incarcerated individuals when the outcomes are still uncertain? This essay argues that if criminal justice system actors choose to use AI, a more responsible place to focus artificial intelligence experimentation may be on improving accountability of the criminal justice system itself. Using AI technology with appropriate governance structures to monitor and improve accountability of criminal justice system actors could provide a breakthrough, addressing the lack of robust mechanisms to provide timely systemic oversight. By starting on systems and system actors instead of the general public or people already caught in the system, AI would be developed in a context where errors of technology development can be mitigated by strong existing protections that have eluded other accountability practices.