A large scale field trial examining the integration of generative artificial intelligence into Pakistan’s justice system found that its use led to an additional 1,848 cases being resolved per year, an increase of more than six percent over the average. The research paper, titled Courts of Tomorrow: Evidence from a Nationwide Rollout of Generative AI, noted that this rise in output did not appear to come at the cost of reduced decision quality.
The study, authored by researchers Sultan Mehmood, Christoph Goessmann and Elliott Ash, examined the deployment of a custom chatbot named JudgeGPT, built on OpenAI’s GPT-4 family of models and adapted specifically for the Pakistani legal context. The tool was tested intensively with the Federal Judicial Academy before being rolled out to 1,559 judges across 118 courts nationwide, covering roughly half of the country’s trial judges and eighty percent of its district courts. According to the findings, the assistant proved most effective when paired with targeted training on how judges should use the tool in their day to day work, with judges who received both access and structured training showing greater willingness to adopt the tool, use it more consistently, and continue relying on it over time.
Judges were divided into three groups for the study, one receiving JudgeGPT access alongside targeted training, another given access with only generic technology training, and a control group that received generic training without any access to the assistant. Researchers tracked outcomes through baseline and follow up surveys on judicial attitudes toward AI, platform usage records showing how judges applied the tool, and district level administrative court data measuring whether AI access translated into higher case resolution. They also reviewed judicial opinions written before and after the trial to assess any changes in writing quality or shifts in language related to gender or religion.
The findings showed that although post trial opinions from judges who used the tool contained more AI generated text than those from the control group, there was little evidence this led to a decline in writing quality, with researchers noting a positive effect on quality assessment if anything. The study also found no meaningful evidence of systematic bias shifts in judicial language following AI adoption. Judges primarily used the tool for legal research and writing support, with targeted training steering usage toward more bounded tasks such as summarization and text improvement rather than full text generation, an approach researchers said helps preserve judicial agency while still improving efficiency. The findings follow earlier national guidelines issued by the National Judicial Policy Making Committee in April, which established a human centered approach ensuring judges remain the ultimate decision makers even as AI tools become more integrated into judicial workflows.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
