Western cultural bias in AI risks shaping billions of users: report

Western cultural bias built into AI systems could influence how billions of users perceive the world as tools enter education, counselling and legal guidance, says a report by Chromatics AI. It notes many AI models are trained on data that underrepresents global languages and worldviews, embedding individualism and Anglo-Saxon norms as default.

The report argues that simple translation or localisation of interfaces is insufficient, as underlying models retain cultural assumptions from original training data. It cites efforts including Egypt's Horus, Nigeria's N-ATLaS, and Indonesia's Sahabat-AI to build culturally grounded systems. Chromatics AI founder Larry Adams said training data must be collected in genuine partnership with communities, and benchmarks must be culturally grounded.

The report warns that as AI takes on more influential roles, embedded values risk shaping under-represented users' worldview in ways at odds with their own culture. The next phase of AI development may depend on how effectively systems understand the cultural contexts of users.