Culturally Grounded Personas in Large Language Models: Characterization and Alignment with Socio-Psychological Value Frameworks

This research investigates the fidelity of Large Language Models (LLMs) in simulating human behavior, specifically focusing on how accurately synthetic, culturally-grounded personas align with established socio-psychological value frameworks. The study aims to conceptualize and generate LLM personas using variables derived from the World Values Survey (WVS) and assess their alignment with the Inglehart-Welzel Cultural Map and Moral Foundations Theory, thereby evaluating their interpretation of stable cultural differences.

Why it matters

The accurate reflection of cultural and moral values by LLM personas is critical for their reliable application in global contexts, impacting areas from strategic communication to policy simulation. Ensuring LLMs can genuinely represent diverse cultural perspectives is foundational for building trustworthy and effective AI systems that avoid bias and promote equitable outcomes.

Key insights

• The study addresses the uncertainty surrounding LLM-generated personas' ability to accurately reflect world and moral value systems across diverse cultural conditionings.
• It proposes the creation of LLM-generated personas based on interpretable variables from the World Values Survey (WVS).
• The research intends to examine these personas through three analytical lenses, including their positioning on the Inglehart-Welzel Cultural Map.
• The primary goal is to determine if synthetic personas accurately reflect established cultural and moral value frameworks.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2601.22396