Does AI help humans make better decisions? One judge's track record—with and without algorithm—surprises researchers
Why did the researchers choose to focus on a single judge's decisions over a 30-month period for their study?
The researchers chose to focus on a single judge's decisions over a 30-month period for their study to ensure a controlled environment for analyzing the impact of AI on human decision-making3. By examining the decisions of one judge, the researchers were able to limit the influence of individual differences among judges and maintain consistency in the decision-making process. This approach allowed them to more accurately assess the effect of AI recommendations on the judge's decisions and compare the performance of the AI system, the judge, and the combination of both in predicting the behavior of arrestees.
What specific factors did the AI algorithm consider when making recommendations about whether to impose cash bail?
The AI algorithm considered age and nine factors related to past criminal experience when making recommendations about whether to impose cash bail. It did not specifically account for race.
How did the AI algorithm perform in comparison to the judge in terms of predicting reoffenders in Dane County, Wisconsin?
The AI algorithm, when used alone, performed worse than the judge in predicting reoffenders in Dane County, Wisconsin. This was determined by analyzing the accuracy of decisions made by a single judge and comparing them to the predictions generated by an AI system, specifically focusing on whether cash bail should be imposed. The AI algorithm was found to be overly harsh in its predictions, suggesting measures that were too severe. However, when the AI algorithm was used to assist the judge's decision-making, there was little to no difference in accuracy compared to the judge making decisions alone.