Kessel OKinga-Koumou was pacing around a crowded hall. She told the crowd of researchers from Africa’s machine learning community that it was her first time presenting. The annual week-long conference (‘Indaba,’ is a Zulu term for gathering) was most recently held in September at Amadou Mbow University in Dakar in Senegal. Over 700 people attended the event to learn about and debate the potential of Africa-centric AI, as well as how it is being deployed in agriculture and other sectors of Africa’s economy.
Okinga-Koumou, a computer science student from the University of the Western Cape, Cape Town, South Africa, spoke about her efforts to solve a common problem at her university: the lack of laboratory equipment. Some lecturers are forced to use 2D printed representations of equipment or chalkboards to simulate practical lessons requiring microscopes, centrifuges or other expensive tools. She lamented that “in some cases, they ask students to draw equipment during practical lessons”.
Okinga Koumou pulled out a phone from her blue jeans pocket and opened a web app prototype she’d built. The app uses AR and AI to allow students to explore 3D models of lab tools in a realistic setting, such as a classroom or laboratory. “Students can have detailed AR of laboratory equipment, which will make their hands-on experiences more effective,” said she.
The Deep Learning Indaba, established in 2017, has chapters in 47 out of 55 African nations. It aims to boost AI across the continent by offering training and resources to African AI Researchers like Okinga Koumou. Africa is still in the early stages of adopting AI technology, but organizers claim that the continent is uniquely hospitable for it due to several factors, including a relatively younger and more educated population, a rapidly expanding ecosystem of AI startups and a large number of potential consumers.
Shakir Mohamed, senior research scientist at Google DeepMind, and cofounder of an organization sponsoring the conference, says that “the building and ownership” of AI solutions tailored for local contexts are crucial for equitable development. Africa is better suited to tackle specific AI challenges than any other continent in the world. Its young talent will also benefit from this, says Mohamed: “There’s amazing expertise all over the continent.” The biggest hurdles are poor infrastructure and inadequate funding. It is expensive to build AI systems. Research to provide AI training in original African languages was also hampered by poor funding of linguistics departments at African universities, and the fact that many Africans do not speak or write their own local languages. Developers may not be able deploy cutting-edge AI technologies due to limited internet access and a lack of domestic data centers.
Complicating matters further, there are no overarching strategies or policies for harnessing AI’s immense benefits – and regulating its downsides. Researchers disagree over a continent wide strategy, despite the existence of several draft policy documents. They disagree on which policies will benefit Africa the most, and not the wealthy Western governments or corporations who have often funded technological innovations. Researchers worry that these issues, taken together, will hold Africa’s AI industry back and hinder its efforts to pave a path in the global AI race.
On cusps of change
Africa’s researchers are already taking advantage of generative AI’s impressive capabilities. Scientists in South Africa have developed an app called Your Choice that is powered by a chatbot based on LLM. This app allows people to share their sexual history with no stigma or discrimination. Kenyan farmers are using AI apps for diagnosing diseases in crops and increasing productivity. Awarri, an AI startup that has just been launched in Nigeria, is working to create the first large language model for the country, with the support of the government. This will allow Nigerian languages to be integrated into AI tools.
Deep Learning Indaba, another sign that Africa’s AI scene is beginning to flourish. Researchers presented 150 posters and published 62 papers at the Dakar meeting. Mohamed says that 30 of them will be published by top-tier journals.
An analysis of 1,646 AI publications between 2013 and 2020 found “a significant rise in publications” by African researchers. And Masakhane, a cousin organization to Deep Learning Indaba that pushes for natural-language-processing research in African languages, has released over 400 open-source models and 20 African-language data sets since it was founded in 2018.
These metrics are a good indicator of the capacity building taking place, says Kathleen Siminyu. She is a computer scientist in Kenya who studies NLP tools in her native Kiswahili. “We’re beginning to see a critical number of people with basic foundational skills. They then specialize.”
Khadija Ba is a Senegalese entrepreneur, investor, and member of the pan-African venture capital fund P1 Ventures. She was present at this year’s conference. She believes that African AI startups are particularly attractive, because their local approaches can be scaled up for the global market. She says that African startups are often forced to build solutions without a robust infrastructure. However, “these innovations work well, making them adaptable for other regions facing similar problems,” she says.
According to a report from the African Private Capital Association, funding for Africa’s tech industry has increased in recent years. VC investments totaled $4.5billion last year, which is more than double of what they were five years ago. Google announced in October that it would be investing $5,8 million to support AI training initiatives across Kenya, Nigeria, South Africa, and other African countries. Researchers say that local funding is still sluggish. Take the Google-backed Fund launched in Nigeria, Africa’s most populous nation, also in October. It will give $6,000 to 10 AI startups, not enough to buy the equipment required to power their systems.
Lilian Wanzare, a lecturer at Maseno University, Kisumu in Kenya, and NLP researcher, is irritated by the lack of support from African governments for local AI initiatives. She also complains that government charges exorbitant amounts for access to data generated by public sources, which hinders data sharing and collaboration. “[We] Researchers are just blocked,” says she. The government says they are willing to support us but the structures haven’t been put in place. There are also major linguistic obstacles.
Ife Adebara a Nigerian computational-linguist asked a question during a discussion at the Indaba: “How many people are able to write a bachelor’s thesis in their own native African language?”
There were no hands raised.
Suddenly, the audience erupted in laughter. Adebara stated that Africans want AIs to speak their native languages, but they themselves cannot write or speak in these languages.
Despite Africa accounting for one-third all languages in the globe, many oral dialects are slowly disappearing as their native speakers decline. LLMs developed in Western tech companies do not serve African languages. They don’t understand the local context and culture.
Adebara, and others researching NLP technologies, believe that the lack of people with the ability to read and/or write in African languages is a major obstacle to the development of AI-enabled technology. “Without literacy in local languages, AI in Africa’s future is not as bright,” she says.
To top it all off, there are few machine-readable data on African languages. Adebara says that the lack of funding for linguistic departments at public universities is one reason. This limits linguists from participating in projects that could produce such data that would benefit AI development.
She and her colleagues founded EqualyzAI this year, a for profit company that aims to preserve African languages using digital technology. They have developed voice tools and AI modeling for 517 African languages.
Lelapa AI is a software company which builds data sets and NLP (natural language processing) tools for African languages. It also tries to address these challenges. The company was founded in 2022 by its cofounders, who met in 2017 at Deep Learning Indaba. In 2023, the company released its first AI tool Vulavula. This speech-to text program recognizes multiple languages spoken in South Africa.
InkubaLM is a small language model, the first of its kind, that supports a variety of African languages, including IsiXhosa and Yoruba. It also supports Swahili, IsiZulu and Hausa. InkubaLM is able to answer questions and perform tasks such as English translation and sentiment analyses. In tests, the model performed as well as larger models. It’s still early days. Jade Abbott is the cofounder and chief operational officer of Lelapa AI. She hopes that InkubaLM, will one day power Vulavula.
Abbott says, “It is the first time we have expressed our long-term vision for what we want and where we envision African AI in the near future.” “What we are really building is a language model that is more powerful than its size.”
InkubaLM was trained using two open-source datasets with 1.9 billion tokens. It was built and curated with the help of Masakhane and African developers working with real people from local communities. They paid native speakers to attend writing workshops in order to collect data for their model. Wanzare says that this approach is fundamentally better because it’s informed and shaped by people who are familiar with the language and culture.
A clash of strategies
Another topic that was brought up repeatedly at the Indaba is that Africa’s AI scene does not have the same regulation and support as other parts of the world, such as Europe, the US and China, or, increasingly, the Middle East.
Only seven African nations (Senegal Egypt Mauritius Rwanda Algeria Nigeria Benin and Benin) have formal AI strategies. Many of these are still in their early stages.
The regulatory framework that will govern AI across the continent was a major point of tension during the Indaba. The African Union Development Agency, which developed this strategy over a period of three years, published a whitepaper in March. The 200-page document contains recommendations for industry codes, benchmarking standards, and a blueprint for AI regulations that African nations can adopt. It is hoped that the heads of African government will endorse it in February 2025, and then the African Union will pass it.
In July, the African Union Commission, a more powerful African governing body than the development agency in Addis Ababa in Ethiopia, released a rival continent-wide AI strategy, a 66-page paper that differs from the original white paper.
The second strategy is not clear, but Seydina NDiaye of the Cheikh hamidou Kane digital university in Dakar, who worked on the white paper for the development agency, says it was written by a Swiss tech lobbyist. The strategy of the African Union commission calls on member states to make AI a priority in their countries, to promote AI startups and to develop regulatory frameworks that address safety and security issues. Ndiaye, however, expressed concern that the document did not reflect the perspectives and aspirations of grassroots African AI communities. “It is a copy and paste of what’s happening outside the continent,” says Ndiaye. Vukosi Marivet, a computer science professor at the University of Pretoria, South Africa, who helped found the Deep Learning Indaba, and is known for being an advocate of the African machine-learning movements, expressed his fury at this turn of events during the conference. “These are things that we shouldn’t be accepting,” he declared. The room was filled with data experts, linguists and international funders. Marivate encouraged them to move forward with the development of AI that would benefit Africans. “We don’t need to wait for rules to act right,” said Marivate. Barbara Glover, program manager at the African Union Development Agency (AUDA), acknowledges that AI researchers have become angry and frustrated. Glover says there has been an effort to harmonize two continental AI strategies. However, the process was fractious. “That engagement did not go as envisioned.” Her organization plans to keep their own version of the continent AI strategy. She says, “We can drive our own AI agenda as Africans.”
INDABA DEEPER LEARNING 2024
All of this speaks to a broader conflict over foreign influence on the African AI scene that goes beyond any one strategic document. Critics say that the Deep Learning Indaba’s reliance on funding by big foreign tech companies is tainted. About 50% of the $500,000 budget comes from international donors, while the rest comes from corporations such as Google DeepMind Apple, Open AI and Meta. They claim that this money could pollute Indaba activities and influence topics and speakers selected for discussion.
But Mohamed says that the Indaba’s cofounder, who is also a researcher at Google DeepMind, “almost all of that money goes back to our recipients across the continent,” adding that the organiza tion helps to connect them with training opportunities in tech firms. He says that the organization benefits from the ties of some of its founders with these companies, but they don’t set the agenda.
Ndiaye claims that funding is needed to keep the conference running. “But we need more African governments to be involved,” he says.
According to Timnit Gebru of the nonprofit Distributed AI Institute (DAIR), who supports equitable AI research for Africa, the concern about foreign funding is a result of skepticism towards profit-driven, exploitative international tech companies. Gebru says that “Africans [need] should do something different, and not repeat the same issues that we’re fighting.” She warns against the pressure to adopt AI for everything in Africa, adding that “international development organizations” are pushing hard to use AI to “antidote all Africa’s problems.” Siminyu is also a DAIR researcher and agrees with this view. She hopes that African government will fund and work together with people in Africa in order to build AI tools which reach underrepresented communities – tools that can be used positively and in a context which works for Africans. She says that “we should be given the dignity of having AI in a manner that others do.”
This story was amended after publication to correct a mispelling of Kessel’s surname and to clarify that the web app she created uses augmented reality technology, not virtual reality.