Study Reveals AI Biases Across Nine Asian Countries

A Singaporean study has uncovered significant biases in artificial intelligence language models across nine Asian nations, including Singapore, India, and Indonesia, Anadolu Agency reported. Conducted with over 300 participants, including 54 experts in linguistics and sociology, the research evaluated models like Sea-Lion, Claude, Aya, and Llama in regional languages and English.

The study identified 3,222 biases categorized into gender, identity, race, socioeconomic status, and unique conditions like caste. For instance, terms like “caregiving” and “teacher” were predominantly associated with women, while “business” and “company” were linked to men. Alarmingly, when asked about online scams, AI models suggested women are more susceptible, and attributed higher crime rates to cities with ethnic enclaves, citing a lack of social cohesion.

Despite cultural differences among the countries, the AI responses showed striking similarities, raising concerns about the implications of biases in AI technologies across diverse populations. The findings call for urgent attention to address these disparities in AI training.