Manners Maketh MAN: Multi-Agent Norm Dynamics under Cultural Moral Values

Keeheon Lee, Kunhee Ryu, Minje Kim, Yoo Ho Gyun

Keywords: Moral Foundations Theory; Norm Dynamics; Multi-Agent Simulation; Large Language Models; Credit Assignment

TL;DR: LLM agents with cultural moral profiles (Japan/US/UK) interact in a shared dormitory under different rule regimes. Rules that complement rather than match a group's culture work best, and mixed-culture groups outperform homogeneous ones.

Abstract: When people from different cultures live together, conflicts often arise—not because of bad intentions, but because shared rules interact with different moral norms. Yet in real-world institutions, it is difficult to systematically test how population composition and rule design jointly shape collective outcomes. We study this question in a controlled setting using an LLM-based multi-agent dormitory simulator. Agents represent culture-conditioned moral preferences drawn from East Asian, North American, and European profiles, grounded in Moral Foundations Theory and equipped with episodic memory. The environment enforces explicit rule regimes (NoRules, Harmony, Liberty, Politeness) through both prompt-level constraints and environment-level penalties. To quantify the gap between individually driven behavior and collective welfare, we introduce a QMIX-based credit assignment module that produces per-agent contribution signals without updating LLM parameters. Across 15 cultural compositions and 4 rule regimes (60 conditions), we evaluate welfare, equity, communication, and credit stability. Contrary to the cultural fit hypothesis, we find that culturally mixed groups under complementary rules outperform homogeneous groups by up to 16.2%, and that rule sensitivity itself constitutes a distinct cultural signature.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 21