At the 9th inaugural Deep Learning Indaba much of the general vibe and conversations revolved around sovereignty–of knowledge, data, technology, and infrastructure. Attending a number of panels, giving my own talk, and having conversations with friends and attendees, ideas of African AI sovereignty were simmering in my head during the course of the conference. What does “African AI Sovereignty" actually mean? What are the practical steps we can take towards it? What are the threats to achieving it in the current AI landscape we are operating in? These questions foreground our workshop, “An Introspective Look at the African AI Landscape” held on the second last day of the conference. In this blog post, I summarize key takeaways from my reflections of the conference as a whole and our workshop specifically.
- Current approaches to “sovereignty” mainly target data, while the issues are dispersed across the entire AI value chain and therefore require the engagement of multiple stakeholders.
A key takeaway from our panel discussion as well as conversations at the end of the workshop and beyond was that the issues around sovereignty are not narrowly concentrated just on “data sovereignty.” For approaches to sovereignty to be meaningful and pragmatic, it needs to include conversations on sovereignty of infrastructure (e.g. compute, data storage, cloud infrastructure), knowledge production and the information ecosystem, policy design and implementation, funding, and economic return to local communities. As workshop participants reflected on the benefits of AI for African communities, a key point of discussion was that the benefits of AI for community members need to materialize as direct, economic benefits. As such, the investment, governance, and distribution of resources needs the involvement and coordination of multiple actors on the private, governmental, and multilateral levels while also actively engaging local communities.
- Depending on their goals, actors in the African AI landscape prioritize private investment for rapid growth vs slow, community centered growth.
Our workshop comprised stakeholders from diverse backgrounds with diverse goals, motivations, and approaches to conceptualizing sovereignty. A key point of contention was on the progress of particularly Asian countries (China and India being prime examples) and how they were able to (seemingly) break from the influence of their colonial past in the area of technological and AI advances. In the discussion that ensued, stakeholders raised three key points 1) the role of government investment in those countries that specifically prioritize AI advances and research, at times disregarding the potential and actual harm to the community and the environment; 2) the need to give “grace” to the African AI community who should prioritize community values and community driven approaches to insure its distinct values and historical roots are embedded in the process instead of comparing with others; and 3) the need to incentivise private African businesses and investors to shift the funding source from its current primary locus (the West) to within the continent. It is important to note that the approaches of the Asian countries mentioned are not necessarily devoid of harm and exploitation to the communities. In fact, panelists clarified the human and environmental cost of some of the initiatives undertaken by these countries and commented that it is not necessarily an approach worth blindly adopting.
- The current approaches of the African AI landscape do not always directly engage African values that are held by community members.
The key question of our workshop was “What is uniquely African about the African AI landscape?” Is it because those working on it are from and/or based in the African continent? Because the work mostly focuses on African languages and datasets? Or because the Deep Learning Indaba conference happens on the continent? Or does work taking place across the African continent adhere to distinctly African values (such as Pan Africanism) that guide AI research and practice? To contemplate on this question, we must first come to consensus around what the community actually considers as African values. From our workshop, three themes of values emerged: Relationality, Cultural Expression, and Communalism. Inclusion of African values, and by extension meaningful inclusion of Africans, thus requires a deeper reflection on how each of those classes of values are integrated in the processes of the several stakeholders in the AI pipeline. In other words, merely including African languages in a large-scale data collection effort is borderline theatrical when considered under the lens of meaningful inclusion of Africa and Africans. Rather, meaningful inclusion requires us to address the uncomfortable yet necessary, social, political, and economic inclusion of community members and the larger value and vision that drives AI research.
- Current criteria for what is considered high-quality training data is largely shaped by the norms of the West, leaving data organically created by Africans as unfit.
Another key takeaway from the workshop and the conference as a whole was on the persistent requirement of the AI landscape for conformity to Western standards, frameworks and norms. Most notable to this was the presentation by Chris Emezue on the African AI Atlas1 which prompts us to reconsider where exactly we look for data. By conducting a broad, holistic search for African data and looking beyond the traditional AI data repositories (e.g. Huggingface), the African AI Atlas project uncovered 7.5 times more data which, under the strict definitions of data and infrastructure, was not quantified. Other conversations also prompted the need to account for the compromises and improvisations that many African countries have to make when “collecting data” due to the constraints on digital tools, power, and internet connectivity. As a result of these compromises, the approaches to data collection in many African communities may thus diverge from the norms of the West. Hence, much of the data “organically” produced by African communities is discarded as unfit as the African AI Atlas project demonstrated. Yet, the conversations at Indaba asked for a rethinking of this. An imaginary where we build for the forms of being as they exist rather than expecting communities to conform to Western norms and standards. I will end this reflection with a quote from Achille Mbembe:
“Every representation of an unstable world cannot automatically be subsumed under the heading “chaos.””
–Achille Mbembe, On the Postcolony, p8
The purpose of the workshop, as its title indicates, was to take an inward look and ask what we are really doing differently and how we are weaving in African values into how we build African AI. I thus leave the reader to ponder on those for key takeaways and ask AI communities across the continent to reflect on how distinctively African values might guide our work and what meaningful inclusion means.
You can read a full report of our workshop with direct quotes and participant inputs here.
Written by: Hellina Hailu Nigatu, PostDoc at AIAL