Digital Disparity: Addressing the Language Gap in AI

AI systems today overwhelmingly favour high-resource languages and datasets, leaving most of the world linguistically excluded. Only roughly 400 languages are fully accessible online, out of circa 7,000 spoken worldwide, constraining training data and services for billions of people. English dominates web content, accounting for around 49% of online content, further skewing the model performance towards other languages. Large-scale initiatives show the problem: industry and telco projects report that many models effectively perform well in only seven high-resource languages, highlighting the urgent need for targeted data, local expertise and funding.  

This session explores practical paths to close the digital-language divide: building diverse, privacy-respecting datasets; partnering with local communities and linguists to capture dialects and context; funding multilingual infrastructure; and creating interoperable standards and open tooling, so models serve everyone. Attendees will leave with a measurable and pragmatic roadmap for turning inclusion from aspiration into operational practice