Colonial Genealogy of the Digital
The standard Silicon Valley self-description presents AI as a radical novelty — a break from everything that came before, ungoverned by the patterns of earlier technologies, accountable only to its own disruptive logic. The colonial-genealogy framework, developed by Mbembe and others including Sylvia Wynter, Ramón Grosfoguel, and Syed Mustafa Ali, rejects this self-description. The patterns by which AI is being built and deployed — the geographic distribution of labor, the concentration of capital in metropolitan centers, the extraction of resources from the periphery, the imposition of governance frameworks written without peripheral input — are not novel patterns. They are the patterns of five centuries of colonial modernity, updated for digital conditions.
The Infrastructural Bargain Underneath — Contrarian ^ Opus
There is a parallel reading that begins not from the persistence of colonial maps but from the material fact that digital infrastructure cannot be built where it does not yet exist. The Democratic Republic of Congo mines cobalt not because Silicon Valley chose to re-colonize it but because cobalt deposits exist there and nowhere else at scale. The periodic table is not a colonial imposition. Kenya hosts content moderation labor not because of metropolitan design but because Nairobi built fiber connectivity, political stability, and English-language education infrastructure that Lagos and Kinshasa have not matched. The asymmetry is real, but its origin is not purely extractive—it reflects compound interest on earlier infrastructure decisions, some colonial, many post-independence.
The epistemic critique assumes that digitized Western text dominates training corpora because of colonial preference, but the actual mechanism is different: text digitization follows from earlier waves of internet adoption, which followed from literacy campaigns, electrification, and archival institutions that many postcolonial states deprioritized in favor of other nation-building needs. The AI training corpus reflects the available digital record, not a curated imperial syllabus. Indigenous knowledge systems remain oral not because AI builders rejected them but because the communities that steward them have—often deliberately, for excellent reasons—chosen not to digitize sacred or sensitive material. The absence is not always extraction; sometimes it is boundary-maintenance by the communities themselves. Recognizing AI's colonial debt is necessary, but overreading continuity risks denying postcolonial states the agency they have actually exercised in building or refusing digital infrastructure on their own terms.
In the AI Story
The argument operates on multiple levels simultaneously. At the material level, the physical infrastructure of AI — the cobalt in the batteries, the lithium in the data centers, the rare earths in the chips — is extracted from the same geographic zones that colonial powers extracted rubber, diamonds, and copper from a century ago. The Democratic Republic of Congo, which produces more than 70 percent of the world's cobalt, is the same Congo that Leopold II turned into a death-world for rubber extraction. The continuity is not metaphorical; it is the same mines, in many cases the same families laboring in them, producing different commodities for different industries.
At the labor level, the outsourcing of content moderation, data labeling, and AI evaluation to Kenya, the Philippines, Colombia, and India follows the geographic logic of the call-center wave of the 1990s and 2000s, which itself followed the logic of manufacturing offshoring, which itself followed the logic of colonial extraction. Each iteration updates the technology; the underlying map remains recognizable.
At the epistemic level — which is where Mbembe's analysis is most distinctive — the training corpora that define what AI models consider knowledge reproduce the hierarchy of knowledge that colonial universities established in the nineteenth century. English-language, Western, male-authored, digitized text is abundantly represented. Oral traditions, minority languages, indigenous knowledge systems are absent or marginal. The model is not neutral; it is trained on a specific intellectual archive, and that archive is the colonial archive.
At the governance level, the regulatory frameworks that attempt to discipline AI — the EU AI Act, the U.S. executive orders, the emerging frameworks in Singapore and Brazil — are written in metropolitan capitals by bureaucrats and technologists who have no accountability to the populations most affected by the technology's deployment. The developer in Lagos is governed by terms she did not write, enforced by courts she cannot access, in languages she may not read. This is colonial governance updated for the platform age.
Recognizing this genealogy is not a counsel of despair. It is a precondition for meaningful intervention. You cannot decolonize what you do not recognize as colonial. Mbembe's work, and the work of the broader decolonial AI movement that has drawn on it, offers a framework for seeing the present clearly enough to imagine alternatives: indigenous training corpora, participatory governance structures, the assertion of African and Asian AI infrastructure on terms that do not require permission from the metropolitan center.
Origin
The colonial-genealogy framework has been developed in dialogue across postcolonial studies, decolonial theory, and critical technology studies. Key contributors include Walter Mignolo, Sylvia Wynter, Ramón Grosfoguel, Syed Mustafa Ali, Shakir Mohamed, Marie-Therese Png, and William Isaac.
Key Ideas
Continuity, not rupture. AI reproduces colonial patterns in updated form rather than constituting a break from them.
Material, labor, epistemic, governance. The continuity operates at multiple levels simultaneously and can only be understood by attending to all of them.
The same geographies recur. The map of AI extraction is substantially the same map as earlier extraction regimes.
Recognition precedes intervention. Decolonizing AI requires first seeing the colonial structures that the industry's self-description denies.
Alternatives are possible. The framework points toward specific forms of resistance and reconstruction, not only critique.
Appears in the Orange Pill Cycle
Stratified Agency Across Levels — Arbitrator ^ Opus
The right weighting depends entirely on which level of analysis you foreground. At the material substrate level—cobalt, lithium, rare earths—the colonial-continuity thesis is approximately 85% correct. The deposits exist where geology placed them, but the terms of extraction, the distribution of profits, the environmental externalities borne locally while value accrues elsewhere: these are structurally continuous with nineteenth-century patterns. The contrarian point about periodic-table necessity holds for 15% of the phenomenon (you cannot mine cobalt where it does not exist), but it does not explain why Congolese miners see 3% of cobalt's value chain.
At the labor level, the weighting shifts to 60/40. Nairobi and Manila host moderation work partly because of colonial-era English diffusion (colonial continuity), but also because postcolonial governments deliberately invested in telecommunications, education, and business-process outsourcing as development strategies (postcolonial agency). Both are true. At the epistemic level, the thesis holds at 70%: training corpora do reflect the digitized colonial archive, but the contrarian point that indigenous communities often chose non-digitization for self-protection is also substantively true and explains perhaps 20% of the absence, with the remaining 10% attributable to technical-linguistic challenges in encoding non-standardized orthographies.
The synthesis the topic itself demands is this: AI infrastructure is over-determined—shaped simultaneously by geological constraint, colonial legacy, postcolonial state capacity, and indigenous refusal. The colonial genealogy is descriptively accurate but explanatorily incomplete. Decolonial intervention requires holding all four forces in view, not collapsing them into a single logic. The map recurs, but the reasons it recurs are plural, and the points of effective leverage differ depending on which mechanism dominates at each node.
Further reading
- Shakir Mohamed, Marie-Therese Png, and William Isaac, 'Decolonial AI,' Philosophy & Technology 33, 2020
- Sylvia Wynter, 'Unsettling the Coloniality of Being/Power/Truth/Freedom,' CR: The New Centennial Review 3(3), 2003
- Walter Mignolo, The Darker Side of Western Modernity (Duke University Press, 2011)
- Syed Mustafa Ali, 'A Brief Introduction to Decolonial Computing,' XRDS 22(4), 2016
- Abeba Birhane, 'Algorithmic Colonization of Africa,' SCRIPTed 17(2), 2020