Machine Learnings

Jail or bail? Machines versus judges

See allHide authors and affiliations

Science  10 Nov 2017:
Vol. 358, Issue 6364, pp. 759-760
DOI: 10.1126/science.358.6364.759-c

Decisions about whether to grant bail could be better made by a machine than by a human.


Predictions based on machine learning could outperform judges when deciding which defendants to jail before trial and which to release on bail. Kleinberg et al. exploited data on more than 758,000 defendants who were arrested in New York City between 2008 and 2013. Compared with carefully devised counterfactual scenarios based on actual judges' decisions, the machine predictions based on defendants' histories could reduce crime by up to 25% with no increase in jailing, or reduce jailing up to 42% with no increase in crime. All categories of crime, including violent crimes, could be reduced, and, critically, so could racial disparities in jailing rates.

Quart. J. Econ. 10.1093/qje/qjx032 (2017).

Navigate This Article