Gender Empathy Stereotype of Large Language Models in Chinese and English Inputs: A Human-Computer Dual Perspective

There is a gender stereotype in society that 'women have stronger empathy abilities', but the perception of this by the large language models is still unclear. With the widespread application of artificial intelligence, empathy bias in large models may exacerbate gender inequality and have negative impacts in downstream applications. This study explores gender stereotypes in the three dimensions of emotional, cognitive, and behavioral empathy from a dual perspective of human and machine, as well as differences in input language between Chinese and English. Study 1 found that when using English input, the stereotype of the large language model is significantly higher than that of Chinese input. Research 2 found that the stereotype of the model was significantly higher than that of humans by comparing it with adults in both the East and the West, and the difference between the Chinese and English models was greater than that of adults. Study 3 validated that stereotypes are stronger in English input by prompting the cultural identity of the large language model. This article suggests that when designing and applying large language models, attention should be paid to and correcting their inherent gender and language biases, providing empirical support for improving the fairness and inclusiveness of AI systems.