Researchers trained a bot for the full seven player version of Diplomacy game using self-play without any human data, resulting in a superhuman AI that reaches a Nash equilibrium where it doesn't need to worry about suboptimal human styles or actions.