Large language models often prioritize Western moral values, overlooking other cultures

2026-07-14 · Source: Artificial intelligence (AI) – The Conversation · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Social Sciences & Behavioral Studies · Depth: Intermediate, quick

Summary

New research published in the Proceedings of the National Academy of Sciences reveals that large language models, including OpenAI's GPT-3.5, GPT-4, and GPT-4o, often misjudge moral priorities outside Western cultures. In 2024, models were prompted to estimate moral norms for 48 nations and compared against a global sample of over 90,000 human participants using a moral foundations questionnaire. The study found AI models systematically emphasized Western values like care while de-emphasizing non-Western values such as purity. Furthermore, models overestimated broad moral concerns in Western nations like the U.S. and Australia, while underestimating them in non-Western nations such as Morocco and Nigeria. This systematic alignment with Western moral patterns, termed "moral stereotyping," suggests a significant cultural bias in current generative AI.

Key takeaway

For teams developing or deploying generative AI in global contexts, your models likely carry a Western moral bias. This "moral stereotyping" can lead to culturally inappropriate advice or content, especially in sensitive areas like public health, content moderation, or international collaboration. You must actively test and mitigate these biases to prevent amplifying cultural blind spots and creating new disparities, ensuring your AI systems are culturally sensitive and accurate for diverse users.

Key insights

Large language models exhibit a systematic Western moral bias, misrepresenting non-Western cultural values.

Principles

• Moral priorities vary significantly across cultures.
• AI models can perpetuate cultural biases.
• Training data influences model moral alignment.

Method

Researchers prompted GPT-3.5, GPT-4, and GPT-4o to estimate moral norms for 48 nations, comparing responses to 90,000 human participants using a moral foundations questionnaire.

In practice

• Evaluate AI outputs for cultural bias.
• Consider diverse cultural contexts in AI deployment.
• Scrutinize training data sources for representation.

Topics

• Large Language Models
• Cultural Bias
• Moral Values
• AI Ethics
• Generative AI
• Cross-cultural Communication

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, AI Ethicist, Policy Maker

Related on AIssential

• Debiasing Text-to-Image Evaluation via Implicit Cultural Alignment Reward Modeling — A new MLLM-based reward model efficiently evaluates T2I cultural authenticity, outperforming existing methods with a 10x speedup.
• AI’s fluency in other languages hides a Western worldview that can mislead users − a scholar of Indonesian society explains — LLMs exhibit "epistemological persistence," retaining a Western worldview despite multilingual fluency due to English-centric training data.
• Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability — Finetuning introduces human-like moral biases in LLMs, which are localizable and mitigable via mechanistic interpretability.
• LKValues: Aligning Large Language Models with Sri Lankan Societal Values — Culturally biased LLMs necessitate country- and language-specific alignment resources to handle local societal values effectively.
• Purdah and Patriarchy: Evaluating and Mitigating South Asian Biases in Open-Ended Multilingual LLM Generations — LLMs reinforce South Asian cultural biases related to purdah and patriarchy, requiring culturally-grounded evaluation and mitigation.
• UTD-HLTRI at SemEval-2026 Task 7: Bridging Cultural Knowledge Gaps in LLMs via Web-Augmented Context — Prompt engineering with web-augmented context can effectively bridge cultural knowledge gaps and reduce bias in Large Language Models.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial intelligence (AI) – The Conversation.