13139: ErnieBot Meets ChatGPT: A Cross-Cultural Examination of AI Biases and Human-AI Creative Collaboration (session 25440)

Existing research has explored cultural biases in Western-developed models such as ChatGPT, but there is limited investigation on non-Western models such as China’s ErnieBot. Additionally, the impact of cultural alignment between humans and AI on creativity and efficiency during collaborative processes, particularly when AI also possesses cultural biases, has not been systematically studied. This paper investigates the cultural biases embedded in large language models (LLMs) and their implications for human-AI collaboration. Study 1 found that AI models trained using data from the United States and China exhibits cultural norms, values, and cognition that are consistent with that culture. Study 2 showed that when human and AI models of the same culture collaborate on local tasks, human experiences greater flow, leading to enhanced usefulness but not enhanced novelty of ideas. These effects did not occur for global tasks. Study 3 replicated Study 2’s findings using field data. Theoretical and practical implications are discussed.

view paper (if available)