Risky Business: Artificial Intelligence and Risk Assessments in Sentencing and Bail Procedures in the United States

30 Pages Posted: 23 Apr 2021 Last revised: 29 Apr 2021

Date Written: December 16, 2020


The American criminal justice system has increasingly looked to predictive analytical technology as a means to predict how likely an individual is to reoffend. The use of actuarial risk assessments is part of a growing trend toward an “evidence-based” approach to criminal justice reform. The application of actuarial risk assessments in the criminal context is significant, because these assessments have the capacity to affect directly the extent to which a state deprives a person of their liberty.

Notwithstanding the widespread support for actuarial risk assessments, some scholars and human rights organizations see the use of these models in the court process as deeply problematic. Beyond questioning the architecture of the algorithm (the examination of which can be severely limited because of the propriety nature of some algorithms), critics also cite concerns relating to the quality of data used by the algorithm, embedded racial bias in the algorithm, and the over-reliance on actuarial risk in sentencing and bail decisions by decision makers who have a limited understanding of the model.

The framework proposed in this paper envisions actuarial risk assessments as a tool for releasing defendants, rather than a tool for imprisonment. The framework delineates an implementation strategy that allows decision makers to benefit from statistical data available to them, while simultaneously circumscribing its use to minimize prejudicial effects to the defendant where issues relating to racial bias and transparency remain unresolved. Important to note is that this framework does not purport to rectify the racial disparity present in the American prison system. Instead, the framework aims not to exacerbate the racial disparity already present and offer an additional avenue for a defendant to argue for release pending trial or for a non-custodial sentence.

Keywords: Artificial Intelligence, Machine Learning, Risk Assessment, Racism, Discrimination, Sentencing, Bail, Criminal Law,

Suggested Citation

Lyn, Alexandra, Risky Business: Artificial Intelligence and Risk Assessments in Sentencing and Bail Procedures in the United States (December 16, 2020). Available at SSRN: https://ssrn.com/abstract=3831441 or http://dx.doi.org/10.2139/ssrn.3831441

Do you have negative results from your research you’d like to share?

Paper statistics

Abstract Views
PlumX Metrics