The cultural stereotype and cultural bias of ChatGPT
Hang Yuan, Zhongyue Che, Yue Zhang, Li Shao, Xianger Yuan, Liqin Huang, Xiaomeng Hu, Kaiping Peng, Siyang Luo
Abstract
Given the rapid advancement of large-scale language models, AI models, like ChatGPT, are playing an increasingly prominent role in human society. However, to ensure that AI models equally benefit human society, we must fully understand the cultural stereotypes and biases that AI models may exhibit in interacting with humans. Study 1 first measured ChatGPT in 13 dimensions of cultural values; it was revealed that ChatGPT's cultural value patterns are dissimilar to those of various countries/regions worldwide. In Study 2, we analyzed the performance of ChatGPT-3.5 and ChatGPT-4.0 across thirteen decision-making tasks involving human interactions from different countries or regions. The results showed that ChatGPT 3.5 exhibits apparent cultural stereotypes in most decision-making tasks and shows significant cultural bias in third-party punishment and ultimatum games. However, ChatGPT 4o, while reinforcing stereotypes, showed reduced cultural bias on these tasks. The findings indicate that, compared to humans, ChatGPT exhibits differential cultural value orientation, and it also shows cultural biases and stereotypes in interpersonal decision-making, and the version upgrade increases the sensitivity of chatbots but weakens the potential bias. Furthermore, Study 3 created four kinds of prompt strategies and applied them to decision-making tasks. It also found that the four prompt strategies can effectively reduce the generation of ChatGPT's stereotypes and biases. Future research should emphasize enhanced technical oversight and augmented transparency in the database and algorithmic training procedures to foster more efficient cross-cultural communication and mitigate social disparities.