Cultural Perspectives and Expectations for Generative AI: A Global Survey Approach

A team of eight researchers, led by Erin van Liemt, has published a foundational study titled 'Cultural Perspectives and Expectations for Generative AI: A Global Survey Approach' on arXiv. The paper addresses a critical gap in AI development: the lack of empirical, global data on how cultures should be represented by generative models like GPT-4, Claude, or Llama. By surveying participants across Europe, the Americas, Asia, and Africa, the team moved beyond theoretical debates to gather concrete community-driven definitions of culture and its desired representation in AI outputs.

The study's core contribution is a set of actionable recommendations for AI developers. It argues that culture must be understood through specific dimensions like religion and tradition, not just broad geographic labels. Crucially, it introduces the concept of cultural 'redlines'—sensitive boundaries that AI should not cross—and proposes a framework for identifying and respecting them. The authors advocate for participatory approaches, suggesting that communities should be directly involved in shaping how AI represents their cultural artifacts, concepts, and values, moving beyond top-down corporate decisions.