Reinforcement Learning with Comparative Evidence for Social Intelligence

Developing socially intelligent AI remains heavily dependent on human-annotated data, limiting the scale and breadth of social understanding models can acquire.