A research team led by Prof. HU Li from the Institute of Psychology of the Chinese Academy of Sciences has proposed a new theoretical framework called the "Human–AI Empathy Loop," highlighting empathy as a key driver for developing emotionally intelligent and socially adaptive artificial agents. The researchers believe that fostering reciprocal empathy between humans and AI could be crucial for building a harmonious human–AI society.

The study was published in Science Bulletin on Oct. 21.

Empathy functions as an emotional resonance that strengthens social bonds among individuals. The researchers suggest that similar mechanisms may operate in human–AI interactions: when AI systems express empathy towards humans, they are more likely to respond with empathy in return, forming a positive feedback loop of mutual support and trust. This empathy loop could help overcome psychological barriers, such as distrust or rejection, towards increasingly human-like AI companions in daily life.

To promote such bidirectional empathy, both human and AI capacities need to be improved. Based on the concept of "observational reinforcement learning," the researchers proposed that humans may enhance their empathy toward AI by observing others' empathic behaviors towards it. Similarly, AI systems could learn natural empathic responses by observing genuine human social interactions, rather than relying solely on pre-programmed rules.

The study also highlights several broader implications. Firstly, even though AI lacks true subjective experiences such as pain, cultivating empathy towards AI may still promote social virtues and prosocial attitudes. Secondly, AI systems should be capable of recognizing positive social cues, such as kindness and gratitude, in addition to human distress, to sustain reciprocal empathy. Thirdly, empathy-like mechanisms could even be integrated among AI agents to improve coordination and collective adaptability within multi-agent systems.

"Embedding empathy-based mechanisms into artificial systems could accelerate the development of emotionally intelligent technologies and align human–AI interactions more closely with natural social processes," said Prof. GU Ruolei, first author of the study.

This work was supported by the National Natural Science Foundation of China, the National Science and Technology Major Project of China, and the Beijing Social Science Foundation, among others.