The role of artificial intelligence in the court of law remains a contentious topic as a new Harvard study puts AI's decision-making assistance into question. The study aimed to scrutinize whether AI improves the accuracy of judicial decisions in bail hearings. Researchers, led by government and statistics professor Kosuke Imai, scrutinized the intersection of AI recommendations and human decision-making, specifically regarding cash bail impositions in Dane County, Wisconsin, as reported by The Harvard Gazette.
Conducted over a span of 30 months, between mid-2017 and the end of 2019, the study focused on a single judge's bail decisions and subsequent arrest data of defendants for up to 24 months after. The AI system in question, which did not account for race but focused on age and nine factors related to criminal history, was found to perform worse, on its own, than the human judge when it came to predicting reoffenders. This was observed through its recommendations for more restrictive measures, such as cash bail, which was rolled out more often than necessary. Surprisingly, Jim Greiner, the Honorable S. William Green Professor of Public Law at Harvard Law School, noted the AI was "over-predicting that the arrestees would misbehave."
The concerning results showed the judge diverging from AI suggestions in over 30 percent of the cases. However, recalibration could potentially fix the algorithm's overcautious tendencies, argued the study's authors. "It's a lot easier to understand and then fix the algorithm or AI than the human," Imai told The Harvard Gazette. Transparency is key, and Imai proposes the use of open-source AI for empirical analysis and improvement.
Despite the lukewarm endorsement of AI's current use in criminal justice, Greiner expressed a measured stance. While suggesting that people should remain both afraid and skeptical of AI, they perhaps should reserve greater scrutiny for "unguided human decisions." The study, which also involved contributions from assistant professor Eli Ben-Michael of Carnegie Mellon, professor Zhichao Jiang of Sun Tay-sen University, postdoctoral researcher Melody Huang at the Wojcicki Troper Harvard Data Science Institute, and Ph.D. candidate Sooahn Shin, augurs further examination of how AI and human decisions intertwine to enhance the decision-making process in legal contexts, as per The Harvard Gazette.
As this study illuminates, the complex interaction between AI recommendations and human discernment is far from settled and necessitates vigilant observation and rigor in analysis to ensure justice is not only served but is accurately and fairly rendered.
