AI
Chat GPT

Africa’s AI future depends on local data, languages and infrastructure

Africa’s most important applications of artificial intelligence are moving beyond chatbots. Nowadays AI is being used to forecast food crises, monitor land use, assess drought conditions and support public health decisions.

But for these systems to deliver meaningful impact, they must work in the realities of the communities they serve. This includes speaking local languages, working with local data and functioning in environments where electricity, devices and computing capacity remain limited.

Microsoft’s 2025 AI Diffusion Report estimates that one in six people globally has used a generative AI product. Yet Africa, home to about 1.5 billion people and more than 1,500 languages, remains poorly represented in many AI systems. This is because they were trained largely on English and a small number of widely used languages.

For a farmer seeking agricultural advice in Dholuo or a mother looking for health information in Amharic, language can determine whether an AI system is useful at all.

Building AI for African languages

There is growing work on closing this gap, much of it being led by organizations and researchers on the continent. LINGUA Africa is an initiative of the Masakhane African Languages Hub, the Gates Foundation, Microsoft AI for Good Lab and Google.org. It supports the development of open datasets, speech resources and language technologies.

Its recent call for projects attracted more than 800 applications from 64 countries, with 85 per cent coming from Africa. The 26 projects selected cover more than 50 African languages across 47 countries.

The projects include work by Arusha Technical College to develop an open, community-led Tanzanian Sign Language resource.  There is also a project by Efficience Globale in Guinea aimed at making vaccination information available in Soussou, Pular and Maninka.

Voice is particularly important in Africa, where oral communication remains central to many communities. Microsoft Research Africa’s Paza project is working to improve speech recognition for low-resource languages. Its benchmark covers 39 African languages, while new models for Swahili and five Kenyan languages have been tested with farmers. This involves using ordinary mobile phones in conditions that include background noise and unreliable connectivity.

Local context matters

An AI system can respond fluently in Kikuyu, Swahili or another African language and still give advice that is unsuitable for the person asking the question.

Agricultural recommendations, for example, need to account for soil conditions, weather patterns, crop cycles and household finances. A technically accurate response can have little practical value if it ignores those realities. This makes access to quality local data just as important as language.

Africa’s data challenges go beyond simply having fewer datasets. Some communities remain underrepresented, maps can be outdated and health records often capture only people who have been able to access healthcare.

AI systems trained on incomplete data risk reproducing those blind spots. The response therefore requires investment in local data collection, documentation and long-term stewardship. This is alongside investment in AI models and computing capacity.

At the same time, African developers cannot afford to wait for perfect datasets or infrastructure. AI solutions need to be designed around the conditions that exist today.

Using AI to address real problems

At Microsoft’s AI for Good Lab, work on food security, public health, environmental monitoring and humanitarian response is being built around the decisions that need to improve rather than the technology itself.

In Kenya, Microsoft has worked with Amref Health Africa and the Ministry of Health to combine routine health records with satellite measurements of vegetation. This is in a bid to forecast acute childhood malnutrition up to six months in advance.

The work is still a proof of concept. Turning such forecasts into healthier outcomes will require deployment, evaluation and time.

AI is also being applied to Kenya’s land-use challenges. Together with the Kenya Space Agency, Microsoft researchers developed land-use maps designed around local landscapes and found that locally trained models could outperform global, one-size-fits-all models.

Other work using satellite imagery to estimate building density and height has been applied to understanding settlement growth, including around a refugee camp in Chad.

When combined with population and cellular coverage data, such information can help identify communities that may be difficult to reach through existing digital infrastructure and early-warning systems.

Kenya’s push for locally owned AI

The emphasis on local ownership is also reflected in ADAPT-Kenya. This is a national initiative led by Kenya’s Ministry of Agriculture and Livestock Development with support from the Gates Foundation and Microsoft’s AI for Good Lab.

The initiative brings together nearly 50 partners to develop AI-powered agricultural data products combining local knowledge with satellite observations.

The applications include crop monitoring, harvest forecasting, market access, insurance and disaster risk reduction. The approach has already been tested during a drought affecting maize-producing counties in Kenya.

ADAPT partners, including NASA Harvest, worked to assess crop conditions and provide decision-makers with a timely picture of the harvest. The partners also plan to apply the approach during the coming El Niño season.

Infrastructure remains a major barrier

Even the most capable AI system will have limited impact if the people who need it cannot access it. Around 600 million people in sub-Saharan Africa still lack electricity, while internet connectivity, access to devices, digital skills and maintenance remain significant challenges.

These issues are sometimes treated as separate from the AI conversation. In Africa, they are central to it. An AI tool that works perfectly in a well-connected research environment may struggle when deployed on a basic mobile phone in an area with intermittent connectivity or where users have limited digital skills.

Africa needs to shape its own AI agenda

Africa is not a single market or a single dataset. The problems faced by a farmer in Kenya, a health worker in Ethiopia and a humanitarian planner in Chad can be very different. AI systems therefore need to reflect local environments rather than applying the same assumptions everywhere.

Grassroots AI communities across the continent have an important role to play alongside universities, technology companies, SMEs, non-profits and international organizations.

Governments will also be critical because they oversee many of the sectors where AI could have the greatest public impact. This includes healthcare, agriculture, education, infrastructure and disaster management.

The goal should therefore go beyond building AI systems that speak African languages. Africa needs AI that is built around its data, understands its local realities. One that works within its infrastructure constraints and earns the trust of the people and institutions expected to use it.

Language is an important part of that equation. But so are good data, reliable infrastructure, local expertise and the ability to turn information into action.

By Dr Girmaw Abebe Tadesse