Plantation Logic

In the AI Story

The plantation was never only an economic form. It was a technology of extraction that required specific cultural and legal arrangements: the dehumanization of the enslaved, the invisibility of the labor to the consumer, the organization of space to separate those who produced from those who consumed. These arrangements did not end with abolition. They were reconfigured, first into sharecropping and convict leasing, then into global commodity chains, and now into the geographic distribution of digital labor.

Content moderation is the paradigmatic example. The AI systems that produce smooth, family-friendly outputs for users in the Global North depend on workers in the Global South who spend eight-hour shifts viewing violent, pornographic, and traumatic content in order to train filters and flag exceptions. These workers are paid a fraction of what equivalent labor commands in San Francisco. They sign non-disclosure agreements that prevent them from discussing what they see. They experience PTSD at rates comparable to combat veterans. And they are rendered invisible in the industry's self-description: the tech company presents itself as having built a safe product, when in fact it has outsourced the exposure to harm.

The same logic structures data labeling, the invisible work of tagging images, transcribing audio, and evaluating model outputs that underlies every large language model in deployment. The labor is geographically distributed to low-wage jurisdictions. The workers have no ownership of the models their labor helps produce. The value they create flows to the metropolitan center. This is plantation logic updated for the digital era — with the crucial difference that the physical plantation made its labor visible to those who walked through it, while the digital plantation makes its labor invisible even to those who depend on it.

The concept is useful not because it claims digital work is identical to chattel slavery — it is not — but because it names the structural continuities that the industry's discourse erases. The smooth interface of the AI assistant is not the product of the engineering team in San Francisco alone. It is the product of a global labor chain whose lower segments remain hidden precisely because visibility would disrupt the consumer experience that the interface is designed to produce.

Origin

Mbembe draws on a tradition of thought about the plantation running from C.L.R. James through Sylvia Wynter, Saidiya Hartman, and Katherine McKittrick. The concept has been extended to digital labor by scholars including Mary Gray and Siddharth Suri in Ghost Work (2019) and Antonio Casilli in Waiting for Robots (2024).

Key Ideas

Invisible labor is the product. The smoothness of the interface is manufactured by making its labor conditions illegible to the user.

Geography is the distribution key. The work is located where wages are low, labor protections are weak, and the workers' visibility to Northern consumers is minimal.

Racial continuity. The demographic composition of the digital plantation tracks the demographic composition of earlier plantation forms with disquieting precision.

The consumer is implicated. Plantation logic works only if those who consume the product refuse to know how it was produced — a refusal the interface actively supports.

Disruption requires visibility. Making the plantation visible is the precondition for organizing against it, which is why plantation logic so aggressively polices what the consumer can see.