Subscribe
Kenya, Nigeria and South Africa top the list of African countries deploying artificial intelligence, according to a new report.
Of the 90 applications identified in the three countries, 40-49 are from Kenya, 30-39 from Nigeria and 20-29 from South Africa, stated the report AI for Africa: Use cases delivering impact.
The report, developed from existing research and interviews with leaders across civil society, non-governmental organisations (NGO), academia and the private sector, focuses on Kenya, Nigeria and South Africa. It also provides regional insights and trends.
Almost half (49 per cent) of the AI applications are in the agriculture sector, followed by climate action and energy with 26 and 24 per cent, respectively. This, according to the report, corresponds to the significant role that agriculture continues to play in Kenya and Nigeria.
The continent’s current contribution to the global AI market valued at $16.5 trillion is only $0.4 trillion (2.5 per cent). AI is projected to increase Africa’s gross domestic product by $2.9 trillion to over $5 trillion.
However, this is not going to be easy if major challenges remain unaddressed. One of the main barriers in Africa is that high-quality and locally relevant data remains limited or difficult to access.
AI has been witnessing major advancements primarily in the Global North, fuelled by the availability of extensive datasets, robust computing resources and models that can be trained efficiently with large volumes of data within shorter timeframes.
These “may not be appropriate or representative for African contexts, and carry inherent risks of exacerbating biases present in the data they are trained on,” reads the report. It called for locally relevant data while also empowering local talent to process and analyse them.
One risk is bias against women. Globally, women are underrepresented in datasets. This is exacerbated by the insufficient collection of gender-disaggregated data in African countries.
Another barrier is the limited language dataset in Africa. Existing large language models (which comprehend and generate human language text like Open AI’s ChatGPT) are primarily trained on data from western and English-speaking countries. And when they are extrapolated to Africa, they can lead to biases and inaccuracies. Local languages are key to ensuring inclusivity and accessibility, the report pointed out.
The other challenge is infrastructure. To train AI, the continent needs high-performance computing, graphics processing units (GPU) and cloud computing systems.
The price of GPU in South Africa and Kenya is nine and 31 times higher than in high-income countries.
African countries also need facilities with storage capacity such as data centres, as well as reliable electricity and the availability of high-speed broadband and mobile internet.
As Africa has limited data centres, its data is often hosted on distant servers and travels through underwater fibre-optic cables. Still, Africa is connected to just a handful of these cables.
“This means that the continent is particularly vulnerable to disruption in these networks, unlike other parts of the world that have robust network redundancy and where traffic can easily be rerouted,” read the report. African countries will need to more than double their data centre hosting capacity by 2030, it added.
Then, there are calls for making data centres energy efficient. Some progress has happened on this front. Data centres such as Amazon Web Services in South Africa and Africa Data Centres in Nigeria, for instance, are exploring renewable energy sources, the authors of the report added. Kenya’s Ecocloud Data Centre became the first African data centre fully powered by geothermal energy.