Continuing from the previous Nature Select team's empathy assessment method, I also reviewed their MAPO training approach. The purpose of this training method is to enable models to maintain a stable "empathetic persona" during multi-turn interactions and provide relatively consistent responses.

In other words, the focus of this methodology is on the behavioral performance of the Agent itself, rather than understanding the true psychological model, guiding human-computer interaction toward a clear "healthy/good" direction.

Furthermore, both papers lack an operationalizable definition of "empathy." In the text, it can roughly be understood as "the ability to demonstrate understanding, support, reassurance, and non-aggressive responses in dialogue." This definition is explicit but clearly not sufficiently deep, and the conceptual ambiguity does not help researchers truly solve the human-computer relational structure issues of Agents.

Original text: