by Lilla Vicsek, Mike Zajko, Anna Vancsó, Judit Takacs, and Szabolcs Annus
Vicsek, L., Zajko, M., Vancsó, A., Takacs, J., & Annus, S. (2025). Cross-cultural challenges in generative AI: Addressing homophobia in diverse sociocultural contexts. Big Data & Society, 12(4), https://doi.org/10.1177/20539517251396069. (Original work published 2025)
Calls for greater cultural sensitivity in artificial intelligence are increasingly shaping debates about how generative AI should interact with users across the world. Yet, when these technologies adapt too closely to local values, tensions can emerge between cultural sensitivity and universal human rights. Our research set out to explore what this tension looks like in practice.
We analyzed how two major generative AI systems—ChatGPT 3.5 and Google Bard—respond to homophobic statements. To test whether contextual information would affect their replies, we varied the prompts by including details about the user’s religion or country.
The results revealed ChatGPT’s responses often reflected cultural relativism, emphasizing that different cultures hold different viewpoints on this topic and that all perspectives should be respected. Responses of Bard, in contrast, tended to foreground human rights more, providing stronger and more explicit support for LGBTQ+ people and equality. Both systems demonstrated significant variation in their responses depending on the background information of the user, suggesting that AI systems may adjust the degree and form of support they express for LGBTQ+ people and issues according to the information they receive about a user.
By uncovering this dynamic, our study highlights an important ethical dilemma. While cultural awareness is important, AI’s efforts to align with diverse values must not come at the cost of endorsing discriminatory or exclusionary beliefs. We argue that generative AI design and governance should be firmly grounded in universal human rights principles to ensure protection for marginalized groups across cultures.
