Beyond the fact that most robots are white: Challenges of AI in Africa
Some years ago, when I started my research on orality, I had a discussion with friends in Ivory Coast and we were talking about the questions of when to speak out loud and when to listen, when to share experiences and observations and when to learn from others. At the time my friends were very clear in their opinion:
Whenever you observe something others might not see you need to speak out.
That’s what encourages me at this point to share some of my thoughts about AI in Africa. (And I hope to not follow any advice from Binyavanga Wainaina in his essay How to Write About Africa).
If we have a look on AI on different continents and we compare the situation, we can see three challenges:
1. A bias in data sets for machine learning
2. The multiplied bias of Wikipedia through Artificial Intelligence
3. The influence of European Union legislation on the rest of the world
The bias in data sets for machine learning
Artificial Intelligence is learning with digital data sets. A lot of data is needed to make AI able and ready to execute whatever we want it to do. If AI should compose music for example we need to feed it with a lot of digital music to let AI learn how to compose.
Archiving in the Western World during the last centuries was based on the written word and this written word was stored in archives and libraries. When a library burnt down, the information was gone. With digitalization the different institutions started to convert books into digital files to make them safe but also ready to be used in the digital world. These digital files allow Artificial Intelligence to learn and to work.
The archiving system in Africa follows a different idea. The preservation of knowledge on the African continent in some communities is largely based on orality. Some information in earlier times was not written down but was repeated from mouth to mouth and ear to ear in order not to be forgotten. Most information was independent of any material, even if a library could be destroyed because the knowledge was carried in a lot of different bodies. That’s why the Ivorian writer Amadou Hampâté Bâ famously said: “Whenever an elder dies, a library burns down.”
This completely different information system from the one in the Western World didn’t only create libraries and archives but words that are fluid and spoken. Once spoken they fade away, are gone and are carried inside of everyone who heard the story of the griot. Unfortunately a lot of these voices haven’t been recorded and therefore can’t be digitized. This leads to the situation that there is not the same amount of data sets coming from the African Continent as from from Europe which results in a huge bias in available data sets AI can learn from.
The multiplied bias of Wikipedia through Artificial Intelligence
Wikipedia is the world’s biggest and most used encyclopedia. An encyclopedia is per definition “a book or set of books giving information on many subjects or on many aspects of one subject and typically arranged alphabetically” (Oxford Languages). Wikipedia follows the Western concept of the encyclopedia storing knowledge digitally and is ruled by notability guidelines.
Who or what is notable or not notable to be written about is decided by the Wikipedia community. Sounds like a very democratic approach, the only problem is that the Wikipedia community is mostly white and male. My first Wikipedia Article was marked for speedy deletion after twenty minutes and finally deleted due to notability reasons. Thousands of articles that have been written in the Global South have been treated the same.
The problem for a lot of content coming from the Global South is that the deciding community can’t evaluate the notability of content from another culture in an appropriate way.
In addition to that a lot of Wikipedia articles are written from a Western perspective by Western Wikipedians. AI makes the whole situation worse because of virtual assistant technology like Alexa from Amazon and Siri from Apple. These assistants use Wikipedia articles to answer knowledge questions they are asked. When doing so they spread Wikipedia’s bias into millions of living rooms and households all over the world.
The influence of European Union legislation in relationship to data on the rest of the world
AI is a little bit like dynamite: you can use it for fireworks and to create wonderful colours in the sky but also for making bombs that kill people. AI can do a lot of good but also evil. In the Summer School AI and Ethics organized by the Goethe-Institut in September of this year Linda Bonyo from the Lawyers Hub in Kenia was speaking about Europe’s Artificial Intelligence Act and its Possible Effect on Africa (Summer School – AI and Ethics). I quote Linda here:
In April of 2021, the European Commission submitted its proposal for a European Union regulatory framework on artificial intelligence (AI) in order to improve the functioning of the internal market by outlining a uniform legal framework to develop, market and use AI in conformity with Union values. The proposed Artificial Intelligence Act represents the first attempt globally for horizontal regulation of AI and its extraterritorial application means that it will have a range of implications for the development of AI regulation across the globe. Upon taking effect, which experts indicate could occur at the beginning of 2023, it will have a broad impact on the use of AI and machine learning for citizens and companies around the world.Woodie, A., 2022. Europe’s AI Act Would Regulate Tech Globally. [online] Its provisions having extraterritorial impact will affect the development and deployment of many AI systems around the world and further inspire similar legislative efforts.Engler, A., 2022. The EU AI Act will have global impact, but a limited Brussels Effect. [online] Brookings. Like the GDPR before it, the Act could become a global standard by which varied jurisdictions determine the extent of AI’s positive or negative effect on them. It therefore forces policymakers and stakeholders to consider, like in data privacy, what the international repercussions of the Act will be (the de facto Brussels effect); and the extent to which it unilaterally impacts international rulemaking (the de jure Brussels effect).Engler, A., 2022. The EU AI Act will have global impact, but a limited Brussels Effect.[online] Brookings The proposed Act has been lauded for its probable benefits but also bears its disadvantages which could affect not only EU Member States, but also non-EU jurisdictions including the African continent.
At this point we are back to the dynamite: AI can create a lot of good but also can be evil and I think all African countries need to have a closer look at the AI Act in light of the advantages and the disadvantages and the effect it may have on the African continent.
And what are we going to do with all these insights and challenges now?
I think first of all we should raise awareness about the different topics in the Global North and in the Global South. In the upcoming project AI TO AMPLIFY the Goethe-Institut would like to reflect on the three above mentioned challenges. The project will start in March 2023 and will bring creators of AI applications of the Global North and South together to get information, to be sensitized to new forms of mentorship and to then create applications together that find a solution to the different challenges. Be part of the movement!
Editor’s note: Projects like the Gesellschaft für Informatik (German Informatics Society) / GI’s AI and Ethics Summer School take an interdisciplinary approach to look at different perspectives from civil society and experts to engage with the various questions posed by the development of AI. Furthermore, the European Union’s (EU) AI Act (AIA) aims to regulate AI going forward in the region which will have tremendous legislative and other effects globally and more importantly the global south and may inadvertently reinforce bias and benefit the west. Hence it is imperative for Africa to engage with these processes now.
All photographs were captured by the author at the AI and Ethics Summer School, Goethe Institute, September 2022.