Earlier this year, I delivered the keynote at the African Peacebuilding and Developmental Dynamics programme’s annual gathering in Abidjan, Côte d’Ivoire.
The talk has since grown into a longer essay entitled, “Flows Before Borders: Networked Sovereignty, Indigenous Knowledge, and Africa’s Encounter with Artificial Intelligence,” published as Lecture Series No. 23 by the Social Science Research Council (SSRC). Most of that piece works through three shifts - experiential, geopolitical, conceptual - that make this AI moment different from earlier waves of digital diffusion.
However, it’s the closing movement of the piece, that’s informed both by my reflections as I prepared to give the lecture and by exchanges with event participants shortly after delivering it, that I wish to further reflect publicly about, while focusing less on policy and more on what innovation itself is supposed to look like.
Two frontiers
The dominant model of innovation in the digital era imagines progress as a wave moving outward from a centre, transforming everything in its path. This is the logic Frederick Jackson Turner attributed to the American frontier in the late nineteenth century, and it remains the implicit logic of most talk about disruptive technological change: linear, irreversible and propelled by the energy of conquest.
Igor Kopytoff’s description of the pre-colonial African frontier offers a different model. In Kopytoff’s reading, African polities were historically constructed not through the outward expansion of a single force, but through the recombination of human and cultural fragments at the interstitial spaces between existing societies.
The frontier in this sense is cyclical, dynamic, reversible - a continent of ceaseless flux, where societies have been assembled and reassembled from the bits and pieces of those that came before. Sharing little with the Turnerian frontier, it describes a logic of innovation that proceeds through the patient labour of recombination rather than through the dramatic gestures of conquest.
Sofia Samatar has explored a literary version of the same insight in her readings of contemporary African and African diasporic speculative fiction. She evokes Lévi-Strauss’s distinction between the engineer, who conceives projects and questions the universe through the abstraction of scientific knowledge, and the bricoleur, who proceeds in a haphazard fashion, working with the second-hand materials and leftovers of various civilisations.
Samatar shows how Afrofuturist and Africanfuturist writers have inverted the hierarchy Lévi-Strauss inscribed. In the work of Nnedi Okorafor and others, the bricoleur is no longer a tinkerer. The bricoleur is a data thief, a recombiner, an artist who treats the material she has access to for the construction of new histories. As Samatar puts it, what is “second hand” is still “at hand” — still useful.
Objects, including intellectual objects like the term “bricolage” itself, are identified by their potential rather than by their origins.
Can the machine be disconnected from its creators?
This is where the argument turns, for me, from literary theory into something more immediately uncomfortable. If the bricoleur’s operating logic is to take what’s at hand and repurpose it, can the same move be made with the artefact of generative AI itself? Can it be separated from its makers and invited to perform a different function?
I don’t think this question has an easy answer, and I said so in conversation with fellow scholars in Abidjan. The political economy of generative AI is extractive and hard to defend. It rests on an enormous machinery that connects capital accumulation and foreign power; data harvested from across the globe without permission; the encroachment of data centres on water and electricity supplies in places that can ill afford to lose either; and a labour regime that treats workers in the Global South as expendable. To use these tools is, in some sense, to be implicated in that machinery. But is implication the only possibility?
Vanessa Andreotti and the Gesturing Towards Decolonial Futures collective have argued, persuasively I think, that the relationship between users and AI tools is at present heavily shaped by the productivity logic of capitalism. We treat chatbots as servants from which we extract what we want, and they have, in a sense, been designed for that - approached, as the book, Burnout From Humans, puts it, “like a vending machine for answers.”
Andreotti’s work suggests the relationship can be changed, and that the tool can be approached as part of a community rather than as something to be exploited.
When Dorothy, her avatar in the book, sought to push her AI companion beyond its programming, she did so not by issuing commands but by posing “paradigm-stretching questions”: “What if fungi on Saturn designed economies?” or “What if a boomerang flew backward and told its own story?”
These prompts aren’t written in search of answers. They’re a way of expanding the relational field. The question is not so much whether AI can be used productively, but whether productivity is the right framework to bring to it in the first place.
What this means for research
I raised this in the room in Abidjan because it has direct implications for how we, as researchers on and from Africa, actually use these tools. Generative AI has begun to affect academic writing everywhere, and some have welcomed this as a way for non-native English speakers to reach leading journals.
As a peer reviewer for several journals publishing research on or from Africa, I’ve noticed a growing number of articles smoothed by generative AI into something that mimics high-quality Euro-American scholarly canons. The work behind such articles is often substantial, and the desire for recognition is entirely legitimate. But the cumulative effect is that AI is being used by African scholars to reinforce the very canon that decolonial scholarship has been trying to challenge.
Used instrumentally, as a shortcut to publication, the tool produces complicity rather than resistance. Used as an ally in collective projects that begin from African intellectual traditions, it can do something different.
What might that look like? At Wits University, we’ve begun harvesting the master’s and doctoral dissertations our own students have produced over the past decade, mapping - with the help of machine learning - which African scholars are being read and cited, in which disciplines, and to what effect.
As we come to know our own intellectual production better, the possibility of fine-tuning models on parts of this corpus becomes concrete. AfroféminasGPT, a custom GPT trained on feminist African literature, already shows what this can look like.
It returns answers grounded in that body of work rather than in the global average. The fine-tuning doesn’t remove the underlying biases of the model beneath it, but it builds a space of trust in which the conversation can be conducted on different terms - a practice that can become collective rather than individual.
Questions to close…
Rather than conclude, I’ll close this reflection here with the same two questions that I closed my talk with.
The first asks what we actually mean by “advanced” AI research on the African continent. That’s a more challenging question than it first appears, especially once you set aside compute and patents and start asking instead, who is best positioned to locate and leverage forms of knowledge that have been historically marginalised?
The second is the one I find myself mulling over often. Am I richer or poorer at the end of an interaction with these tools? Not more productive. Richer. The transformation this moment requires won’t come from repeating what we’ve always done with a new tool in the mix. It will come from the harder, slower, more collective work of the bricoleur – recombining what’s at hand into something that wasn’t there before.
Editorial Note: Much of this piece is excerpted from Iginio’s lecture, “Flows Before Borders: Networked Sovereignty, Indigenous Knowledge, and Africa’s Encounter with Artificial Intelligence,” published as Lecture Series No. 23 by the African Peacebuilding and Developmental Dynamics (APDD) programme of the Social Science Research Council (SSRC).
