Developing accurate and representative multilingual language models remains a major challenge as AI systems expand globally. In particular, differences in resources, data availability, and cultural context in the Global South can make it difficult to build models that perform consistently and fairly across the world’s thousands of languages. While Large Language Models (LLMs) have advanced rapidly in English-speaking regions, most of the world’s 7,000+ languages—especially low-resource ones—remain underrepresented, with only 5–10% included in advanced language technologies. This leads to significant performance gaps and limited AI safety measures for non-English languages.
This session convenes RAI Fellows to discuss practical approaches for building non-English AI systems and advancing culturally aware AI deployment.