ASPIRING TO ELIMINATE ARTIFICIAL INTELLIGENCE BIAS IN AFRICA

News broke that Google was opening the first AI Center in Ghana and it is going to be their first in Africa. In a blog post Jeff Dean Google AI fellow and staff research scientist Moustapha Cisse wrote “In recent years, we’ve … witnessed an increasing interest in machine learning research across the continent,” and this development has piqued a lot of interest around the world.

AI’s exceptional growth and impressive advancements are not limited to specific geographies but rather have an impact on all continents, Africa included. However, many African countries are still battling with issues related to the first, second and third industrial revolutions such as electricity, mechanization of production and automation. Therefore, questions about Africa’s preparedness for the fourth industrial revolution are being raised: Is Africa catching up with the continual advancement in technology? From cheap abundant labor to natural resources, Africa’s current strengths seem not to match with the fundamental needs of the fourth industrial revolution that consist mainly of colossal investment capital, research and development (R&D) and highly-skilled talent. However, the ongoing industrial revolution represents an opportunity, if used well that will enable Africa to become a main player in the world economy.

However, Africa has found a way to use technology in the best way possible not letting the limitation of lack of regular electricity and strong internet connection among other things hinder the increasing interest of Africans to machine learning and other technological developments.

According to the Encyclopedia of Human Behaviour, the term Cognitive bias was introduced in the 1970’s by Amos Tversky and Daniel Kahneman, who defined it to mean

“A systematic error in judgment and decision making common to all human beings which can be due to cognitive limitations, motivational factors and/or adaptations to natural environments”.

A number of cognitive biases have been discovered and there are at least 51 different biases that fall into the thinking and judgment categories. Algorithmic bias has become a topic of attention in the societal conversations on machine learning and artificial intelligence. In the past months, several instances of algorithmic bias have made the news — such as misclassification of humans by image recognition algorithms.

Algorithms play a key role in the functioning of autonomous systems and so concerns have periodically been raised about the possibility of algorithm bias. They play a critical role in all computational systems most especially autonomous one. The possibility of algorithm bias is worrisome for auto and semi-autonomous systems, they usually don’t involve humans in the loop who can detect and compensate for biases in the algorithm model. Equally importantly, these different biases for algorithms can arise from many different sources. We thus turn to the task of disentangling different ways in which an algorithm can come to be biased.

Danks and London in their research stated the taxonomy of algorithm bias to include the following:

Training data bias, Algorithm Focus bias, Algorithm Process Bias, Transfer context bias and Interpretation Bias. The research also concluded by stating that are many different types of algorithmic bias that arise due to multiple classes of standards or norms, and from many different sources.

Africa is a vast continent with over 135 ethnic groups, and over 2000 languages spoken in Africa and a population of over 1 billion people. The extent of diversity means that in the development of AI diversity must be considered and bias should be eliminated as much as possible, it is almost illogical to think that all ethnic groups and languages can be represented but the vastness of AI has proved that it is quite capable of handling it.

A couple of weeks past, I watched a video by Joy Buolimni  (Click here ) where she showed how AI systems evaluated the faces of famous black women in the world and the system read their faces to be that of men, giving them title of “male, square faced” etc. She mentioned that this among others is one the many reasons why AI bias should be consciously on the mind of the people who are creating this systems.

It dawned on me that if popular black women can be profiled wrongly what will happen when it involves Africans living in Africa with the many differences would most likely be classified to be the same. As AI starts to grow in Africa, especially with the fact the Google AI research has come to Africa, Ghana precisely there is a need to have as much as possible a diverse team with more Africans than Caucasians especially if we are building AI to be used in Africa. A perfect example of bias has to be the word embedding, a popular algorithm used to process and analyze large amounts of data which usually characterizes European American names as pleasant and African ones as unpleasant, or Microsoft Word that always underlines African names to mean wrongly spelt but never the same for European or American names.

As more technology companies are delving into AI research and activities, it is important to involve Africans who are capable of working on this project to ensure diversity. Devices, programs and processes shape our attitudes, behaviors and culture. AI is altering economies and societies, changing the way we communicate and work as well as restructuring governance and politics. Our societies have long endured inequalities. AI should not inadvertently sustain or even worsen them, as technology serves to make the lives of men better.