Generative AI (GenAI) – specifically, language-based models – has received global attention since the launch of OpenAI’s ChatGPT in November 2022. The technology has transformative potential in Africa, where it could democratise access to information and services at the last mile.
This blog explores why GenAI has so much potential for impact, explains three challenges to deploying GenAI in Africa and offers four recommendations to support its growth across the continent.
Why is GenAI so revolutionary?
Based on user prompts, GenAI models leverage natural language processing (NLP) and empower users to interact with the technology on their own terms without needing advanced technical skills. This significantly lowers barriers to entry and expands access, especially among marginalised groups.
Yet despite its potential, the development and adoption of GenAI in Africa remains nascent. Our recent report, ‘AI for Africa: Use cases delivering impact’ shows that an overwhelming majority of use case applications in Kenya, Nigeria and South Africa rely on predictive AI, a more established form of AI that identifies patterns from historical data and predicts future trends.
Source: GSMA Mobile for Development
Barriers to the development of GenAI in Africa
Lack of locally relevant data
One of the main barriers to the development of GenAI in Africa is the lack of locally relevant data, particularly local language data. Adoption of digital services largely depends on their relevance, which requires offering solutions in local languages to ensure inclusivity and accessibility, especially at the last mile.
Source: Mozilla Foundation
The availability of local language data is critical for GenAI solutions that rely on NLP yet only 0.02% of online content is in African languages compared to 53% in English. This considerably limits the ability to create new language models or fine-tune existing ones to reflect the diverse range of accents, dialects and ways of speaking across different African communities. Promising initiatives are emerging, such as Awarri in Nigeria, Lelapa.ai in South Africa, and Mozilla Foundation’s Common Voice project – but their efforts remain limited in scope.
Running costs
GenAI solutions have significant computational requirements compared to predictive AI due to their complex architecture and reliance on large datasets. They require large clusters of graphic processing units (GPUs) with specialised chips capable of processing huge volumes of data across billions of parameters. This brings two key challenges.
First, the affordability of GPUs is a significant barrier in Africa. While absolute costs are similar across regions, they are disproportionately high in Africa income levels. In Kenya for instance, the price of a GPU represents 75% of GDP per capita, making it relatively 30 times more expensive than in Germany.
Second, GenAI requires a substantial amount of electricity, due to the energy needs of GPUs. A search on a large language model (LLM) may consume ten times the electricity of a Google search. This is particularly problematic in Sub-Saharan Africa, where countries are already struggling to cope with power demand.
Skills gap
Countries and companies in Africa have a limited talent pool available due to factors such as a persistent skills gap and significant brain drain. Few local start-ups have the resources and talent to build local language models. This trend is somewhat mitigated by the increasing availability of open-source LLMs, which local AI developers can fine-tune and deploy within existing applications – providing they get access to sufficient locally relevant data.
As most AI-enabled services are delivered via mobile channels, end users do not need technical skills but do require some level of digital literacy to interact with GenAI. They need to be able to carefully craft input prompts, referred to as prompt-engineering, to get the best results.
Source: GSMA Mobile for Development
How to support the development of GenAI in Africa?
Enhance local language data availability
Dedicated funding should be allocated to entities that specialise in bridging the language data gap, including those leveraging community-driven data collection methods. Raising awareness on the unique requirements – in terms of time, resources and expertise – of building language datasets and models will be essential to unlock funding. There are also untapped opportunities to leverage existing data, for example (anonymised and aggregated) call detail records from mobile operators.
Leverage partnerships
Securing partnerships – and, accessing funding – is critical for local actors who may otherwise not have the required resources and capabilities to develop GenAI solutions. For example, Digital Green partnered with OpenAI to get access to its technology suite, and with CGIAR and the Kenya Agricultural and Livestock Research Organisation (KALRO) to get access to locally relevant secondary data to develop Farmer.Chat.
For more examples of local use cases, read our recent report ‘AI for Africa: Use cases delivering impact’.
Support access to compute and promote renewable energy
There is a need to address challenges such as power outages and to conduct energy audits in data centres to help enhance operational efficiency and foster long-term infrastructure resilience. Countries in the region are less reliant on fossil fuel than many others globally, and have a distinctive opportunity to champion innovative, green data-centre infrastructure and pioneer new approaches to clean computing. This will be essential to balance growing energy demands of the population with the energy requirements of data centres. Governments should assess their compute needs to establish roadmaps for targeted investment in infrastructure.
Upskill end-users
As GenAI gradually emerges across Africa, raising awareness and offering training on prompt-engineering will be critical to ensure users can learn how to formulate effective prompts that result in desired outcomes, in turn building trust and confidence in GenAI. Prompt-engineering training can be incorporated into existing digital literacy and skills programmes, such as the GSMA’s Mobile Internet Skills Training Toolkit.
Want more? Read about over 90 AI-enabled use cases tackling development challenges across agriculture and food security, energy, healthcare and climate action in our AI for Africa report series here.
The Central Insights Unit is currently funded by the UK Foreign, Commonwealth & Development Office and supported by the GSMA and its members.
