A study conducted by researchers from ETH Zurich, Imperial College London, and the New Economic School found that an AI assistant called JudgeGPT significantly improved judicial productivity in Pakistan. The tool, built on OpenAI's GPT-4, was designed to assist judges in trial courts by retrieving relevant legal documents and generating cited answers. The experiment involved 1,559 judges across 118 courts, with results showing a measurable impact on case resolution rates. The researchers split judges into three groups, with two receiving AI access and one serving as a control. The group that received targeted training with JudgeGPT saw a substantial increase in case resolution compared to those with minimal training or no access. The study highlights the potential of AI to enhance productivity in public sector institutions when paired with proper training.

The AI system, JudgeGPT, was used to search a database of 129,235 legal documents, including court rulings and laws. When a judge entered a query, the tool retrieved the ten most relevant passages and generated a cited response. Trained judges used JudgeGPT four times as often as those in the general seminar group, with the trained group averaging nearly 60 logins and over 200 prompts. The control group had about 20 logins and fewer than 50 prompts. The study also found that trained judges resolved more cases, with districts having more trained judges clearing about 1,848 extra cases per year per district, a 6.3 percent increase. Even in the lowest-performing districts, the number of resolved cases rose by about 616.

The researchers emphasized that the AI system did not replace judges but supported them by handling routine tasks. The study found no increase in gender or religious bias in judicial language, and ruling quality either held steady or improved. A review of 4,000 court judgments showed a slight improvement in rulings from trained judges, with 59 percent of pairwise comparisons rating them better than those from the control group. The findings suggest that AI can enhance public sector productivity when users are properly trained, but without such training, most gains disappear.

Source: thedecoder