Appreciate the pushback, but I think this misreads the mechanism. Mediator.ai doesn't decide; it generates candidate agreements, scores them against both sides' stated preferences, and presents the best one. Either party can reject the proposed agreement. The parties still have to agree. That's facilitation, not arbitration.

On the hidden-interests point: the assistant actually tries to tease out unstated preferences. That's what the conversation with each party is for, and it uses several preference-elicitation strategies to get at what's underneath a stated position - but I'm sure there is plenty of opportunity for refinement here.