What happened
A field study showed a 6,3% increase in case resolution among users trained on the JudgeGPT system, while the effect disappeared without instruction.
Why it matters
The results demonstrate a critical dependence of artificial intelligence effectiveness on the quality of user preparation, casting doubt on automation strategies that ignore the factor of human training.
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.
Facts
- 1559 Pakistani judges participated in the experiment.
- Use of the AI assistant JudgeGPT increased case resolution by 6,3%.
- A positive effect was observed only among judges who underwent practical training.
- Without training, the effect of using the system practically disappeared.
- The estimated return on investment is up to 38,50 for every dollar invested.
Context
Information is based on a report by The Decoder regarding a field experiment conducted in Pakistan. The data represents a meta-description of a study published in July 2026.
What remains unknown
- What was the duration and specific methodology of the practical training for the judges?
- Which specific types of legal cases showed the greatest increase in resolution speed?
- Is this training model scalable to other jurisdictions with different levels of digital literacy?
AI analysis
The high return on investment ($38,50 per $1) likely includes not only the software cost but also the social impact of accelerating justice. The disappearance of the effect without training indicates that the interface or logic of JudgeGPT is not intuitive without prior user adaptation, creating a barrier to spontaneous technology adoption in conservative institutions.
Strategic AI conclusion
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.