In a field experiment involving 1559 Pakistani judges, it was found that using the AI assistant JudgeGPT led to a 6,3 percent increase in the number of resolved cases. However, the key condition for success was the presence of practical training: efficiency gains were observed exclusively among judges who underwent special instruction on how to use the system.
Researchers have no data on positive effects in groups that did not receive training; in such cases, the tool's effectiveness effectively dropped to zero. The study authors estimate the economic return from implementing this technology, accounting for training costs, at up to 38,50 for every dollar invested.
The findings underscore that the mere deployment of algorithms does not guarantee solutions to judicial system problems. Without investing resources in human capital and staff adaptation, technological innovations may prove useless for reducing court backlogs.
editorial commentary
Why it matters
The most likely consequence will be a revision of court digitization budgets to increase allocations for staff training. The next observable signal will be pilot projects in other countries focusing specifically on training methodologies rather than software procurement. The main uncertainty lies in the long-term retention of skills by judges after the training program ends.