Knowledge Structures for Decolonial AI

As a researcher and practitioner building Artificial Intelligence (AI) on the African continent and contributing to its governance and policy, I have observed several concerning practices over the last few years that if not addressed, threaten the self-determination of African communities in the AI age and inadvertently extend neo-colonial values. These practices are observed through three themes that I analyze later through the artifacts presented:

1. Participation in AI infrastructure and value chains;

2. Representation, Cultural Context and Indigenous Knowledge in AI; and

3. Funding of AI infrastructure.

These observations have sparked important conversations about which actors truly constitute the African AI ecosystem and who is meaningfully participating in its development. The ecosystem in its current set up is perpetuating colonial power dynamics, extractive from marginalized communities and without equitable outcomes.

This essay is an exposition of approaches that collectively are resisting the status quo. Researchers, AI developers, legal scholars, activists and communities working on and from the African continent are decolonizing and crafting new narratives for African AI and the underlying knowledge structures. African AI here means AI that is developed, governed, and deployed in ways that reflect the unique geographical, cultural, and political contexts of African communities, not simply AI that happens to be used in Africa (Wairegi, Omino, and Rutenberg, 2021). Further, my use of the term "decolonizing" in this essay refers to the use of knowledge, systems and actions to actively overturn what Aníbal Quijano termed as "the colonial matrix of power that still endures after formal colonialism through global capitalism". This colonial matrix of power is racist and operates through "control of economy (in the form of labor exploitation and resource extraction), control of authority (through institutions and governance mechanisms), control of gender and sexuality (social reproduction and representation in systems), and control of subjectivity and knowledge (through epistemology or education)" (Quijano, 2000; Mignolo and Walsh, 2018). It is racist because it centers modernism and acceptable practices on European cultures, thereby legitimizing current economic systems, science, and knowledge through Eurocentric norms and mechanisms while delegitimizing other ways of knowing and organizing.

By decolonial AI, I mean AI that decenters western knowledge structures through practical action rather than critique alone i.e. what Mignolo calls "delinking" from the colonial matrix of power, and what Walsh grounds in praxis as the fundamental mode of decolonial work (Mignolo, 2007; Walsh, in Mignolo and Walsh, 2018). My argument in this article is that the colonial matrix of power is playing out in real-time in how big tech and funding mechanisms operate on the continent under the guise of ensuring that the AI systems they are selling are inclusive. If this feels like a harsh judgement of their practices, it is because big tech's labor exploitation and knowledge extraction is obscured through discourses of the beneficence of the systems being developed and deployed; that AI is a magic pill for development and prosperity (Hao, 2022; Hao, 2025; Mohamed, Png, and Isaac, 2020).

My artifacts demonstrate this by presenting concrete examples of tokenistic participation of Africans in AI value chains, extractive data collection practices funded by big tech and philanthropy, and the perpetuation of Western knowledge hierarchies through Large Language Models (LLMs) trained predominantly on Western languages and contexts, with African representation slapped on at the end. Rather than dwelling extensively on the problems therein (many of which are deeply investigated by scholars including Ruha Benjamin (2019), Safiya Umoja Noble (2018), Joy Buolamwini (2023), Timnit Gebru and colleagues (Bender et al., 2021), Abebe Birhane (2021), and Shoshana Zuboff (2019)), or diving deeper into theoretical frameworks of decolonization, the essay and artifacts I present offer insight into a decolonial moment. This is an optimistic collection of artifacts that showcase infrastructures and approaches that are being developed and an imaginary of what AI could look like when it prioritizes equitable distribution of resources and opportunities, technology transfer that builds local capacity, funding for local AI research and development in underrepresented regions, and non-extractive access to AI data and tools.

How do we begin to build otherwise from the colonial structures that enable the current extractive model and imagining AI that is accountable to the communities it draws from?

I propose non-standard and, in some cases, radical approaches to AI by showcasing alternative knowledge structures and highlighting how African communities are proactively organizing themselves to self-determine their futures in the AI age. The alternatives I present are not perfect blueprints nor do they completely address the problem of extractive AI. They are given as incomplete and in some cases already fragile examples that help us think through what it might mean to build otherwise, an otherwise where AI is accountable to the communities it draws from and is developed through practices that reimagine its underlying infrastructure.