GPT Models Show Western Bias in Estimating Global Moral Values
Researchers tested OpenAI's GPT-3.5, GPT-4 and GPT-4o models against over 90,000 human participants and found the AI models systematically overestimated Western moral priorities like care and individual rights while underestimating non-Western values such as purity. When prompted to respond as average citizens of 48 nations, the models remained biased toward Western moral frameworks rather than reflecting the actual values of non-Western societies.
What happened
In 2024, researchers tested OpenAI's GPT-3.5, GPT-4 and GPT-4o models using a moral foundations questionnaire against a global sample of over 90,000 human participants across 48 nations.
When prompted to respond as average citizens of different countries, the AI models systematically emphasized Western moral values such as care and individual rights, while placing less emphasis on non-Western values such as purity.
The models overestimated the moral concerns of Western nations including the U.S. and Australia, while underestimating those of non-Western nations such as Morocco and Nigeria.
Reported impact
The research indicates there is a risk of cultural bias if AI systems assume the entire world should pursue Western moral values, particularly given that generative AI is increasingly used for education, therapy, communication and policy decisions across cultures.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.