Algorithms can outperform humans
To put it to the test, Tan and his colleagues constructed an original dataset of 50 “high-volume, low-value” conflicts—disputes that occur frequently but are too small for the courtroom, such as bungled deliveries, office conflicts and landlord-tenant disagreements.
They fed these scenarios to a suite of LLMs and instructed them to analyze the situation, select the best type of intervention and draft an appropriate message. Human mediators were asked to do the same. Blind evaluators then judged each response. Was the approach sound? The message clear, neutral and on-point? Did it fully grasp the situation?
Overall, the AI models delivered striking performances, surpassing humans in some respects.
In 62 per cent of cases, the LLMs came up with an intervention strategy that matched or surpassed the human-generated one. And in 84 per cent of cases, the AI-generated mediation messages were rated as good as, or better than, those crafted by humans. The LLMs also scored high on criteria such as impartiality, understanding and contextualization.
A tool for wider access to mediation
Tan believes these results open up new possibilities for mediation. First of all, the use of AI could expand access to mediation for minor conflicts, everyday disagreements and online disputes.
“AI could deliver immediate, anonymous help at a fraction of the cost—without having to wait for an appointment,” he said.
These technologies could also support human mediators on online dispute-resolution platforms such as Parle Consommation, without replacing them.
“AI can deftly handle and quickly summarize simple cases, freeing up time and resources for professional mediators,” Tan said. “On the other hand, for sensitive or emotionally charged conflicts—those that require empathy, judgment or managing messy human dynamics—caution is still called for.”
In essence, AI isn’t yet poised to replace human beings, but it can provide an accessible, fast and neutral first level of assistance.
