A nationwide experiment integrating generative AI into Pakistan's trial courts has yielded a measurable increase in judicial productivity—without compromising the quality of rulings. The findings, published Tuesday in a research paper titled Courts of Tomorrow: Evidence from a Nationwide Rollout of Generative AI , show that judges using a custom AI assistant resolved an additional 1,848 cases per year, a 6.3% increase over the average. Researchers Sultan Mehmood, Christoph Goessmann, and Elliott Ash designed the study to test whether AI could help address chronic case backlogs while preserving judicial independence. Main Developments The field trial involved 1,559 judges across 118 courts—roughly half of the country's trial judges and 80% of district courts. These judges were randomly divided into three groups: one received the AI assistant—named JudgeGPT, a chatbot based on OpenAI's GPT-4 family—with targeted training on judicial applications; a second group got the tool with only generic technology training; and a control group received generic training without access to the assistant. JudgeGPT was customised for the Pakistani legal context and beta-tested intensively with the Federal Judicial Academy before deployment. The trial measured outcomes through baseline surveys of judges' attitudes toward AI, follow-up surveys three months after rollout, platform usage records, district-level administrative case resolution data, and analysis of judicial opinions written before and after the trial. Read also: Proclamation Proceedings Launched Against KP CM Afridi in Protest Case Background Pakistan's judiciary has long grappled with a backlog of cases, a challenge common to many developing legal systems. In April, the National Judicial Policy Making Committee (NJPMC) formally issued national guidelines for AI use in judicial institutions, emphasising a human-centric approach where AI assists but does not replace judicial decision-making. The guidelines promote ethical and transparent use, with safeguards against bias and an emphasis on explainability and accountability. The study builds on earlier research into AI's potential in legal settings, but this is among the first large-scale field trials to measure real-world impact on case resolution rates and judicial writing quality. The researchers note that sustained engagement with the tool depended strongly on targeted training, not just access to the technology. Why It Matters For judiciaries facing persistent backlogs, the findings suggest AI can be a practical tool for improving state capacity—but not a cure-all. The study found that targeted training shifted judges' use of the tool toward tasks best suited to language models, such as text improvement and summarisation, rather than full-text generation or open-ended legal queries. This approach preserved judicial agency while boosting productivity. The research also found no evidence that AI use reduced writing quality or introduced systematic bias in judicial language related to gender or religion. "If anything, there is a positive effect of AI on the quality assessment," the paper notes. Judges who received both AI access and targeted training were more likely to adopt the tool, use it more intensively, and continue using it over time, with their attitudes shifting toward expecting productivity gains. What's Next The NJPMC's April guidelines provide a framework for broader AI adoption in Pakistan's courts, but the study's authors caution that AI is "not a panacea." The paper concludes that when the tool is built around relevant legal materials and paired with training that directs use toward appropriate tasks, it can become a practical tool for improving state capacity. Further research may explore whether similar results hold in other jurisdictions or with different AI models.