After 60 minutes of deliberation from leaders of top AI communities in Africa, we learnt that the single major challenge blocking Africa’s trained data scientists from using available data to build world class solutions is the difficulty and expense in accessing fast computers, built to make AI a reality in other continents.
I was excited to learn of the Alliance4ai community. I gained amazing insights at their first inaugural gathering in June, and then decided to sign up as a volunteer to be a part of the team. To say it was exuberating is an understatement; from being part of the organizers, planning the second gathering of AI community leaders, to sharing experiences and meeting innovative minds in solving some of Africa’s greatest challenges in AI.
The event took place on the 8th August 2020. It commenced with presentations from two brilliant AI leaders – videos here and here – followed by breakout sessions. I coordinated the breakout room session on compute access.
This section focused on some of the top challenges facing accessing compute in AI in Africa; we A diverse group of AI community leaders from Ghana, Nigeria, Kenya, Tanzania Morocco and others talked about nuances in their societies and suggested solutions.
Computational power or compute in AI
With oceans of data to understand, only powerful compute power, mostly delivered by GPUs, allow researchers to unlock golden insights in a reasonable time, at feasible costs. Using the wrong compute can increase your model training time from 2 days to 2 months, and multiply your costs significantly.
Challenges with compute
Top challenges discussed in my room session were;
- Low access to GPUs; GPUs are designed to massively accelerate the speed of training AI models, and solving groundbreaking problems. Dina Machuve, an AI community leader of the Lacuna Fund explained that there is a lack of speed in processing data in Africa and added that cloud networks and data bundles are expensive.
- Storage capacity is a challenge; the growing amounts of data collected in Africa begs for tailored solutions that are affordable and allow for quick access of the data. ~ Fabian Fawole, Alliance4ai, Nigeria.
- Two participants, Eugene Oduma from Kenya and Israel from Ghana, both AI leaders stated that there is also the issue of individuals not knowing what GPUs to select for their desired purpose.
Other problems raised during the discussion were;
- Insufficient research done on AI computing
- Lack of experts in maintaining AI compute resources with low awareness
- Collection of complex data or data which is not numerated that will require advanced processors to process data.
Possible solutions to the above challenges were addressed by the community leaders as;
- Dina Machuve, an AI community leader of the Lacuna Fund, Tanzania suggested that work should be done with governments to update policies that make it incredibly expensive to import GPUs to African countries. In the meantime, people should write guides and tips on how they were able to purchase and build their own GPU clusters
- Allocating more funding and encouraging computing research. Universities should invest in high computing sensors. ~ Alfred Ongere, AI community, Kenya.
- Open source projects like ‘Kaggle and Google Colab’ should be promoted to practitioners ~ Eugene Oduma, AI community, Kenya
- Rose Delilah Gesich from Nairobi Women in Machine Learning and Data Science and Israel from Zlitch technologies, Ghana, shared similar thoughts by suggesting experts build or deploy software to orchestrate the sharing of available compute resources to communities across the continent.
We ended the discussion with a good sense of satisfaction and I enjoyed the conversation with all the community leaders. They hoped to join the next edition of Alliance4ai leaders gathering.
The question however remains, what can AI leaders in Africa do today to provide compute access for their communities?
Do you wish to provide solutions to these challenges or state any pressing challenge relating to this, then you are welcome to join the Alliance4ai team or attend the next gathering if you lead an AI community in your country. Complete this form to be considered for an invite to the next gathering.
Continue the conversation by tagging us on twitter @alliance4ai and use the hashtag #alliance4ai_leaders.
About the Author
Janet Akorfa is an Environmental science student at the Kwame Nkrumah University of science of Technology and a beginner in website design and development. She is the co director of the Yielding Accomplished African Women for the university chapter, an organization that empowers women in finance and technology. She has served as a volunteer for Yalilearns, KNUST, and well known as a student’s activist.