Cultural tendencies in generative AI

Summary: Peer-reviewed study finding generative AI systems reflect and amplify dominant cultural norms, raising concerns about representation and language loss for minority cultures.

Relevance: Relevant to the Ethics & Environment subdomain (representation and cultural/linguistic bias): shows generative AI outputs shift by the language a user prompts in, with implications for social workers using AI translation or content tools with non-English-speaking or culturally distinct clients.

Key findings: Testing GPT (and replicating with China’s ERNIE model) across established cultural-psychology measures found GPT is more interdependent/collectivist and holistic in Chinese versus more independent/individualist and analytical in English; in a downstream test, GPT prompted in Chinese was more likely to recommend advertisements with an interdependent social orientation than the same model prompted in English, though the study examined only two cultural constructs.

Read the original: https://www.nature.com/articles/s41562-025-02242-1