Africa Is Expanding Its Reach In Artificial Intelligence

South Africa will lead the way with the first phase of the accelerated computing set, to be launched in June. While specific timelines for the other countries remain unannounced, Cassava Technologies has committed to delivering robust AI computing power that promises to boost productivity for African businesses and researchers alike.

In a LinkedIn post, Zimbabwean entrepreneur Strive Masiyiwa, founder of Cassava Technologies, said the vision behind this initiative is to ensure Africa remains part of the fourth industrial revolution. He highlighted the critical importance of "direct access to what we call 'compute' — the capacity to train AI models and process prompts from within Africa, and use the output to create our own intelligence ecosystem."

Recognizing African contributions to global AI

This development is happening at a time when Africa's contributions to AI development have often been overlooked. Tech workers in Kenya, Uganda, and Nigeria have been essential to the AI pipeline, providing data training, content moderation, large language models (LLMs), and other crucial services for tech giants like OpenAI and Meta.

Justin Irabor, an AI Engineer at Germany's International University of Applied Sciences, sees Cassava's initiative as necessary and overdue. "Africa has not just been a consumer of technology when it comes to AI; it has been a contributor," he tells OkayAfrica. "Africans are the reason LLMs sound the way they do, which means that LLMs carry the DNA of Africans. Obviously, you want more control over the means of production."

Irabor points to the political dimensions of AI technology, "The internet technically is value-neutral, but LLMs have meanings and political backings to them. So, it makes no sense for the continent planning to assert itself in the global space to sit back and not have its own opinions on all these technologies. It could be potentially hazardous if Africans don't take a stand."

The language bias challenge

Last year's controversy surrounding ChatGPT's language patterns highlighted these concerns. American computer scientist Paul Graham claimed that words like "delve" were telltale signs of "text written by ChatGPT." Many Africans criticized this as fostering biases against original writing, while Westerners expanded the list to include words such as "demystify," "safeguard," "harness," and "empower."

In her 2020 paperAlgorithmic Colonization of Africa, Mozilla's AI Accountability Senior Advisor Abeba Birhane warned, "The importing of AI tools made in the West by Western technologists may not only be irrelevant and harmful due to lack of transferability from one context to another but also is an obstacle that hinders the development of local products."

Creating African AI infrastructure

Olubayo Adekanmbi, CEO of EqualyzAI, sees AI as a critical factor in national competitiveness. He argues that building products using African contexts is essential to prevent digital colonialism. "AI is powered based on what it learns, so it learns by the quality of the data it is fed," Adekanmbi tells OkayAfrica. "ChatGPT, Mistral AI, Anthropic, Gemini, and others are scraping internet data. But unfortunately, 95 percent of conversation style, linguistic diversity, and nuances of Africa, whether in the form of speeches, text, images, and videos, are not on the internet."

Adekanmbi advocates for small language models (SLMs) to better capture contextual data for African content but notes that data sovereignty remains a significant challenge that Cassava's new data centers aim to address. "GPUs help AI engineers organize that knowledge. But if we continue relying on the big players, our data will not be on the continent," he says. "This means that anybody building an SLM will have to move the data collected – terabytes of audio, video, and image – and send it to a data farm in Ohio or somewhere in the US where they have GPUs. There might be some latency because it's very far."

The advantages of local infrastructure are clear. "What if we have those GPU capabilities on the continent? This means our data will stay on the continent. And the fact that local businesses power the data means it will be cheaper for us. Meanwhile, as a small business in Africa, I can just rent a small space within the Cassava AI infrastructure to test something or build a model for agriculture, climate, health, or education. And pay as you go. The implication of that is that our developers can innovate fast."

Environmental considerations

Despite the innovative nature of AI, there has also been growing concern about its environmental impact, particularly the considerable electricity and water consumption required for operation.

David Akinmade, machine learning researcher and founder of Laroye AI, sees these challenges as opportunities for innovation. If harnessed correctly, AI would provide solutions to these environmental concerns. "This is the advantage of AI; it can equip you with the resources that allow you to set up in a more efficient, less environmentally impactful way," he explains to OkayAfrica. "There are lots of studies we've done around how we can use the initial models built to smaller models that perform at almost the same level, or slightly lower, for a fraction of the energy."

Akinmade is optimistic about the broader impact, "If we do it well, the amount of resources required to get that same level of intelligence would reduce. Africa shouldn't be bothered by what it will do to our climate. If we do it right, it will help us figure out ways to avert or do our best to be more prepared for what is coming in terms of climate change. Building data factories, maintaining them, and unlocking new levels of economic productivity is a great thing for Africa, provided those factories are built using local companies that hire local talent."