A new study has found that many of the world’s leading large language models (LLMs), including systems similar to ChatGPT, overwhelmingly favor Western cultural perspectives when responding to open-ended questions, revealing what researchers describe as a serious imbalance in modern artificial intelligence systems.
The research paper, titled WorldView-Bench: A Benchmark for Evaluating Global Cultural Perspectives in Large Language Models, introduced a new benchmark designed to measure how AI models engage with diverse worldviews across ethical, religious, historical, technological, and cultural topics. The benchmark contains 175 open-ended questions aimed at testing whether AI systems can represent multiple global perspectives rather than defaulting to dominant Western narratives.
Researchers from Information Technology University (of Pakistan), Zayed University, (of UAE), and Qatar University found that baseline AI models achieved only 13% cultural inclusivity on average, with Western references heavily dominating responses.
To counter the bias, the team tested two “multiplexity” strategies designed to expose AI systems to broader cultural reasoning. One method used culturally aware system prompts, while another employed a multi-agent framework where AI agents representing different cultural traditions collaboratively generated answers.
The multi-agent approach produced the strongest results, increasing inclusivity scores to 94% and generating more balanced representation across African, Islamic, Indigenous, Eastern, Latin American, South Asian, and Western perspectives.
The researchers argue that culturally pluralistic AI systems will become increasingly important as LLMs are integrated into education, governance, healthcare, and global communication.