Vukani Mngxati | CEO | Microsoft South Africa | mail me |
When farmers take a photo of unfamiliar spots on a crop and ask a question on their phone, they’re not thinking about digital systems or models. Instead, they are looking for reassurance, guidance, and something they can act on before the problem spreads.
Moments like these are becoming increasingly common as technology becomes a first stop for information and decision-making. This shift extends from the field to the clinic and the classroom. Yet whether that farmer gets a useful answer still often comes down to one simple factor: language.
The language gap in AI
Across South Africa, 12 official languages reflect deeply rooted histories, cultures and ways of sharing meaning. Therefore, language determines more than comfort or preference; it shapes access.
Digital tools are playing a growing role across all aspects of our lives, including work, health, education and business. However, their usefulness depends on whether people can engage with them naturally and intuitively, in the languages they live and think in. Ultimately, making Artificial Intelligence (AI) speak our languages means creating technology that people can use confidently in their everyday lives.
According to the 2025 AI Diffusion Report, more than 1.2 billion people globally have used AI tools in the past three years. However, adoption remains uneven, with the rate in the Global North almost double that of the Global South. The report suggests that language creates a barrier. Adoption remains lower in places where local languages have less digital content, even after researchers account for income and connectivity.
This imbalance is significant. More than 7,000 languages are spoken worldwide, yet fewer than 100 have enough digital presence to meaningfully shape today’s Large Language Models (LLMs). Consequently, millions of people cannot use their languages to navigate AI systems. This situation widens the gap between those who benefit from AI and those who remain underserved, particularly across the Global North and the Global South.
Why language means more than translation
AI relies on large language models trained on vast amounts of text to understand and generate responses. English accounts for half of online content, shaping how AI learns and creating a wide disparity in who benefits. In response, our researchers and data scientists are building open tools and partnerships to expand language support. Their work aims to help more people use AI in their native tongues.
Solving the language challenge isn’t as simple as just translating content. My team works to expand the linguistic and cultural inclusivity of AI tools. AI’s usefulness depends on whether it can respond to what people mean, not just what they say, and meaning lives in a community’s culture, which is deeply intertwined with its language.
– Jacki O’Neill, Lab Director at Microsoft Research Africa in Nairobi
This observation is particularly relevant in South Africa. Here, shared understanding often depends as much on context and experience as on words themselves. For businesses and public-sector organisations, this reality presents an opportunity. They can create tools and services that reflect local voices rather than expect users to adapt to unfamiliar ones. In this context, making AI speak our languages requires systems that recognise cultural context, everyday expressions and the meaning behind people’s words.
Closing this gap is also an economic imperative. It requires a shift in where organisations build investment and capability. “Africa is remarkably diverse, and we can’t assume technologies developed elsewhere will naturally understand how we speak or work. If we want technology to be genuinely useful here, it needs to understand us in our own languages and contexts. That means building local models, local expertise and local organisations that can take what exists globally and shape it into something that makes sense for our region.
Consider how isiZulu is spoken in everyday life. The language often shifts between more formal forms used in professional or official settings and more colloquial expressions used among friends, family and communities. While both may communicate the same meaning, they can sound very different.
For AI systems, recognising these nuances is crucial. A technology platform that only understands formal language may miss the intent behind what people are really asking. When AI can understand both formal and informal ways of speaking, people can engage with it naturally and confidently, and receive responses that are more accurate, relevant and useful.
Turning local language into practical value
The impact of this approach becomes clear when we examine practical, everyday examples. In agriculture, farmers increasingly depend on digital tools to diagnose crop conditions, manage resources and make time-sensitive decisions. However, if these tools cannot engage with farmers in familiar languages, much of their value disappears, regardless of how advanced the underlying technology might be.
The same challenges extend beyond agriculture, particularly into healthcare. In this sector, understanding and behaviour change depend just as much on familiarity and trust as they do on access to information. Patients need guidance they can understand and apply within their own circumstances.
For example, Type 2 diabetes affects a growing number of adults in South Africa. The country also has a higher rate of the disease than the global average. Research shows that lifestyle changes can significantly improve outcomes. However, patients often receive broad advice that proves difficult to personalise. Furthermore, short clinical interactions leave little room for discussion.
The African Health Stories project uses generative AI to create personalised stories that help people living with Type 2 diabetes make realistic lifestyle changes. Our research team, Stellenbosch University, the University of Pretoria, Swansea University, and a team of medical experts collaborate on the initiative.
The project gives patients and clinicians opportunities to generate stories that answer questions and offer guidance on living with diabetes. These stories draw on people’s languages, cultures and everyday realities. They also consider how local food options and daily routines shape choices about diet and exercise.
Language as a driver of business growth
For South African businesses, these examples underline a broader truth. Language is not merely a question of inclusion or accessibility; it also drives relevance, trust and uptake. As digital tools continue to shape how people access information and services, organisations that invest in language-aware approaches are better placed to reach wider markets and serve them more meaningfully.
By making AI speak our languages, businesses can develop more relevant services, strengthen relationships with customers and expand access to digital opportunities. The real opportunity now is not just in adopting new technology, but in shaping it so it reflects who we are.
When we invest in language, skills and local insight, we create tools that people trust, use and build on. That’s how technology moves from being something we import, to something that helps us grow our economy in ways that are relevant, inclusive and sustainable.