Cultural Value Alignment in Large Language Models: A Prompt-based Analysis of Schwartz Values in Gemini, ChatGPT, and DeepSeek

LLMs embed cultural values rather than universal ethics: The study found that DeepSeek, trained on Chinese-language data, prioritizes collectivist values (downplaying individual achievement and power) while Western models like ChatGPT and Gemini show different patterns, demonstrating that AI systems reflect the cultural biases of their training data rather than operating from a neutral ethical stance.

All major LLMs converge on prosocial values: Across Gemini, ChatGPT, and DeepSeek, self-transcendence values like benevolence and universalism were highly prioritized, suggesting LLMs are generally designed to emphasize helping others and the common good—though this apparent consensus may mask deeper cultural differences in how these values are interpreted.

Current AI alignment approaches may inadvertently impose Western values globally: The research highlights that without intentional design choices, LLMs risk spreading particular cultural value systems worldwide, raising concerns about fairness and cultural neutrality in AI systems used across diverse populations with different moral frameworks.

Addressing value misalignment requires new technical approaches: The authors propose three specific strategies—multi-perspective reasoning, self-reflective feedback, and dynamic contextualization—to help LLMs better accommodate diverse cultural values rather than defaulting to a single value system.

One-size-fits-all AI ethics is insufficient: The findings underscore that responsible AI development requires "pluralistic alignment frameworks" that intentionally integrate multiple moral and cultural perspectives, rather than assuming a universal ethical framework can work for all users globally.