The study conducted by Oxford University and published in the Quarterly Journal of Economics suggests that using algorithms to make some judicial decisions could result in fairer judgments for defendants. By relying on machine learning-based models, companies can eliminate some of the systemic biases that judges may possess. These models are driven by assumptions about human behavior, which are used in various applications such as product recommendations on Amazon and predictive texts on phones.
Researchers analyzed New York’s pretrial system and found that a significant number of judges make systematic prediction errors based on defendant characteristics like ethnicity, age, and prior behavior. The study used data from 1,460,462 cases in New York City, with 758,027 cases subject to pretrial release decisions. By developing statistical tests, researchers identified judges who made systematic errors in prediction and provided methods for estimating biased predictions.
The study revealed that at least 20% of New York judges exhibited systematic errors in predicting defendant’s risk of misconduct. Scientists estimated the effects of replacing judges with algorithmic decision rules and showed improvements of up to 20% in trial outcomes. However, the impact of replacing human judges with algorithms depends on various factors such as the political class’s objective and the trade-off between human prediction errors and private information.
According to Ashesh Rambachan, the main author of the study, when considering factors like ethnicity, age, and the type of crime charged, judges make systematic prediction errors for approximately 30% of defendants. Replacing judges with algorithms has the potential for better outcomes in trial results measured by lower failure-to-appear rates and pre-trial detention rates. It is crucial to evaluate the trade-offs between human errors and private information to determine the effects of algorithmic decision rules in the judicial system.
