What key choices do you answer when you build AI solutions, and deploy them to users in Africa?
This was the key question of the breakout room I led during the August 8th 2020 second bi-monthly gathering of Africa AI community leaders, organized by Alliance4ai.
Prof Khalil’s presentation highlighted the key strategies and responsibilities of AI communities and recommended a novel AI strategy which is a way forward for most AI communities in Africa.
For data scientists, researchers and social entrepreneurs who are interested in data collection and gathering, the Lacuna fund is giving out funding to get data from the underserved population in the fields of Agriculture, Health and Language. Learn more about the eligibility of the Lacuna Fund from Dina’s presentation.
Ethics, the big issue…
Together with some AI community leaders, we delved into the issue of Ethics in AI especially in the African setting.
AI has always been and is expected to do more and eventually automate every aspect of the human routine. However, we forget to see that in real life, humans tend to make sudden critical decisions that have not been thought of before to either save lives or a situation. Every society and community is bound by some ethical rules and guidelines which cannot be compromised. Having AI’s act like humans, can we then entrust all the aspects of automation to an AI and be sure of making only the best out of it? Ethics and fairness has been characterized by what we deem right or wrong in the society and as such needs to be highly considered in our quest to explore the uses of AI.
Data bias has been a major cause of unintended flaws in most AI applications. This leads to under-representation of marginalized communities and it continues to have an effect on the output of AI systems that are built.
“I really feel the underrepresentation of these marginalized communities on these systems, which I think is a serious problem now. Such systems, we use the same systems, we use Facebook, Google yet we are not sure about the representation of Africa”- Dina Machuve, Data Science Africa
Another problem is that our choice of where to apply AI can lead to large swings of job gains or job loss. Rose Delilah Geischo, a learning program intern at Alliance4AI, shared a view that some tasks should not be managed by AI simply not because it will have greater negative effects on many lives.
“So now we have lots of banks in Kenya lending funds to the middle or low income. There is an issue of low education on how these systems work. You get rejected for a loan and yet do not know the reason because it is an AI that is trained to review applications“ – Eugene, AI Kenya
“Discrepancies that come with the training data not as a result of not collecting right data. But at the end of the day, the data set has been manipulated to achieve the expectation of the people who are implementing the design” – Israel Adjei-Yeboah, Zlitch Technologies Limited.
Would we want to hold AI developers accountable for the failure of the system other than the AI itself?
The way forward
Fabian Fowole (Alliance4AI), stressed on the fact that inasmuch as there is the fear that AI would eventually get people out of jobs, it is undeniable to note that AI improves productivity and efficiency in most industries despite its threats.
Biases based on empathy and status quo are not consciously implemented but could be unavoidable. Rather than allowing AI to rule every aspect of life, why don’t we decide on what and what not the capabilities of an AI should be? In this case, it is very necessary to have a laid-out documented guideline to limit the design by developers which intend to place a restriction on the capabilities of the AI.
Identifying what the uses of AI are, with justifiable reasons on why a particular AI should be implemented needs to be highly considered.
“The challenge of AI ethics is one of choice – what lives should AI improve, what actions should AI be allowed to take and what conditions need to be met before these actions are allowed in your society. Let’s build a framework together that will define these parameters”. – Alex Tsado, Alliance4AI
Next, making available an open data repository to gather and store datasets especially African data or the marginalized, creates a great platform for having authentic balanced data which duly represents Africans. “Having open data access is some form of solution to identifying some of the biases in data…”- Israel Adjei Yeboah
Education and understanding in algorithm design and concept of ethics should be a key area of interest to the individual communities. “Aggressiveness to access professional training in AI for the underrepresented. AI education at all levels from primary school” – Dina Machuve. Educators and researchers should find a credible source of information concerning ethics to guide its implementation.
We believe AI has great potential to create an economically-driven society when properly exploited and regulated. Alliance4AI will look into coordinating work around gathering informed opinions on tasks AI should be used for, and key questions AI builders should ask themselves as they build for Africa.
For more on the other breakout room discussion, read Janet Akorfa’s blog on “Addressing the Nature of Compute in AI”
If you would love to contribute to these discussions please follow us on social media @alliance4ai on Twitter, LinkedIn and Instagram with the hashtag #alliance4ai_leaders and complete this form for an invite to the next gathering.
About the Author
Martha Teye is an MSc Computer Science student at Kwame Nkrumah University of Science and Technology. She is also a Software Developer at Zlitch Technologies Limited, Accra. Martha is the co-founder of Tech Flair, an initiative that seeks to train Senior High School students with programming skills in Ghana. She has research interests in Machine Learning and the implementation of Artificial Intelligence in the educational sector.