Value Diversity Proposed as System-Level Metric for Multicultural AI Agents

Evaluating 19 cultures and 18 backbone models on the World Values Survey, this work shows that per-agent cultural alignment scores are largely uncorrelated with system-level value diversity across multicultural multi-agent deployments. The paper argues alignment metrics miss whether a multi-agent system collectively preserves cultural plurality.