Continuity through transformation
The journey from diamond mines to data mines is neither a clean break with the past nor a repetition without change. Digital economies extend familiar relationships of appropriation and concentrated control. They also transform what can be appropriated, how people participate, and where commercial value is produced.
I call this proposed new phase Extractive 2.0. The term is not a validated historical periodization. It is an interpretation of a consequential shift: digital infrastructures can make everyday participation a continuing source of commercial value while retaining dependence on mineral extraction, energy, and human labor.
Extraction and extractivism are not interchangeable. Extraction concerns processes through which resources are obtained. I use extractivism to describe arrangements in which powerful actors secure disproportionate control over resources and accumulation while displacing costs onto people with limited authority over those arrangements. Extractive 2.0 names the transformation and expansion of these relationships through contemporary digital economies.
Neither resource use nor data collection is sufficient, by itself, to establish extractivism.
The question is how appropriation is organized: who sets the terms, who benefits, who bears consequences, and who can meaningfully challenge the arrangement.
The mineral-system pattern
Mineral systems provide more than a metaphor for this inquiry. They make visible relationships that the finished product often conceals.
A mineral moves through extraction, trading, processing, manufacturing, consumption, and disposal. Its changing commercial value does not automatically become greater authority or well-being for the community from which it originated. Understanding that relationship requires looking beyond the mine: at ownership, purchasing terms, processing capacity, labor protections, and the institutions governing access to benefits.
The Berlin Conference of 1884–1885 did not begin African resource extraction or colonial exploitation. It helped establish rules through which colonial powers recognized territorial claims, without consulting African representatives. Its significance lies partly in that institutional arrangement: external powers negotiated authority over places and populations excluded from the negotiations. [web:52]
Contemporary mining cannot simply be read as an unchanged colonial relationship. But the historical question remains useful: whose authority makes extraction possible, and whose consent can be bypassed?
Consumption and disposal belong within the same inquiry. The digital device does not cease to be material when it reaches its user. The Global E-waste Monitor 2024 estimated that 62 million tonnes of e-waste were generated worldwide in 2022; only 22.3% was documented as formally collected and recycled. It also identified substantial uncontrolled, undocumented movements from high-income to middle- and low-income countries. [web:42][web:43]
These figures do not mean that all discarded electronics are exported, or that African countries are merely disposal grounds. They show why technological progress cannot be assessed only at the point of consumption. Its material trajectories continue after the screen goes dark.
From diamond mines to data mines
My “diamond mines to data mines” thesis concerns a changing organization of value, not the claim that data is another mineral.
Nick Couldry and Ulises A. Mejias describe “data colonialism” as the appropriation of social life through data relations. Their argument helps explain why data is not simply a resource waiting to be collected: infrastructures and institutions organize activities so that they become available for commercial processing. [web:32]
What changes in Extractive 2.0 is the relationship between consumption and ongoing production. In data-intensive services, using the product can generate information that supports targeting, prediction, or improvement. Participation becomes part of the resource stream. This differs from merely selling a finished object, although not every service collects or commercializes participation in the same way. [web:32][web:27]
Attention introduces another relationship. In advertising-supported arrangements, sustained engagement can create more opportunities for advertising and behavioral measurement. Users receive real services, but platform operators control important conditions of access and visibility. The extractivist concern arises where commercial appropriation expands while participants have little influence over its terms, subsequent uses, or consequences. [web:32]
Paid digital labor is different again. Data labelers and content moderators perform work under employment or contracting arrangements. Their wages, exposure, protections, and bargaining power cannot be understood simply by calling every platform user a worker.
A specific case makes these distinctions concrete. Beginning in November 2021, OpenAI sent text to the outsourcing firm Sama for labeling in Kenya. The work included classifying graphic descriptions of violence and sexual abuse for systems intended to detect harmful content. TIME’s January 2023 investigation reported take-home earnings of approximately $1.32–$2 per hour, based on documents, payslips, and interviews. The four workers interviewed described psychological harm. [web:25]
OpenAI commissioned the work and controlled the downstream AI system; Sama employed the workers and managed their working conditions. TIME reported a contractual rate of $12.50 per hour paid to Sama. That rate was not a worker’s wage, and the difference cannot simply be labeled profit: Sama said it covered infrastructure, salaries, benefits, supervision, and quality assurance. [web:25]
The reporting also documented disagreements. Sama disputed the reported workload and wage range and said professional counseling was available. OpenAI said Sama managed pay and mental-health provision, and that its understanding included workers’ ability to opt out without penalty. Those responses do not erase workers’ accounts; they reveal why responsibility and enforceable protection must be examined across the contracting relationship. [web:25]
Here, commercial value moved toward a deployable AI system, while exposure was assigned through an outsourced labor arrangement. The question is not whether safety work was necessary. It is whether the people performing it had adequate protection and meaningful authority over its conditions.
A comparative anatomy of extraction
The following table compares relationships within extractivist arrangements. It does not imply that all mining or digital activity belongs in those arrangements.
Analytical concern
Mineral-centered arrangements
Digital-centered arrangements
Core resource
Minerals and materials entering production
Data, recorded activity, and attention; supported by labor and material infrastructure
Sites and infrastructure
Mines, trading networks, processing facilities, factories, and disposal sites
Platforms, devices, workplaces, data centers, and their physical supply chains
Labor and participation
Extraction, processing, manufacturing, and recovery work; conditions vary
Paid annotation, moderation, and technical work; user participation is a separate relationship
Externalities and sacrifice
Physical injury, pollution, ecosystem damage, and displaced livelihood costs
Psychological exposure, privacy loss, and constrained autonomy, alongside physical environmental burdens
Beneficiaries, control, and value capture
Distributed among firms, intermediaries, states, and workers; communities may have limited decision rights
Distributed among developers, platform owners, contractors, and workers; contributors may lack authority over subsequent uses
Justification and terms
Development, employment, and growth; contested through rights, contracts, and regulation
Innovation, convenience, and access; governed through contracts, platform rules, and data practices
This is an analytical synthesis, not empirical validation of equivalence. Its shared concern is the separation of control and accumulation from exposure and responsibility. Its limits are equally important.
A depleted ore body is not copied data.
Mining injury is not privacy loss. Paid employment is not ordinary platform use. And digital infrastructures remain physically situated, however distant they appear from the mine. [web:27][web:32][web:25]
The proposed new phase lies in the conjunction: participation can continuously replenish commercial inputs, infrastructures can organize that participation, and the resulting systems remain dependent on older material and labor relationships. No single feature proves a historical rupture. Together, they justify investigating a transformation rather than treating digital extraction as either immaterial novelty or an unchanged past.
Extractive Obesity, sacrifice, and participation
In the original formulation, I used Extractive Obesity to describe excessive exploitation of resources and labor from mining communities. I now clarify the orientation behind that excess: an insatiable appetite for more resources, production, consumption, accumulation, and potentially technological capability. Extractive Obesity concerns a worldview driving expansion, not a diagnosis of individuals or a scientific measure.
Extractive Overshoot names a possible condition in which demands associated with that orientation exceed identifiable ecological, social, human, or institutional limits.
Obesity can generate pressures that lead to Overshoot; Overshoot does not produce Obesity, and the relationship is not inevitable.
These are my working conceptual formulations. Establishing overshoot in a particular case would require identifying the limit, whose assessment counts, and the evidence that demands have exceeded it.
Sacrifice and privilege clarify what is otherwise hidden within expansion. In the Kenyan labor case, protection for downstream users depended in part on workers confronting harmful material. I describe the risk of a “mental sacrifice zone” where psychological exposure is concentrated among people with limited influence over the system their work enables. This is an interpretation of the relationship, not a clinical diagnosis or a claim that all annotation work is traumatic. [web:25]
Privilege is not simply the absence of harm. It can be a position within a system that delivers convenience or protection while keeping its conditions distant. The same person may benefit in one relationship and be disadvantaged in another.
Our lives have been choreographed around technological consumption. That choreography is neither total nor irresistible. But products establish defaults, invite repetition, and make some actions easier than others. Where participation in work or social life depends on a platform, clicking “agree” does not resolve whether the terms are meaningfully negotiable. Constrained choice is not identical to the coercion faced by a miner or employee; it is another relationship requiring scrutiny. [web:32]
The Expanding Appetite for Intelligence
Could the pursuit of increasingly capable AI express this orientation toward continual expansion? The question is not whether greater capability is inherently undesirable. It is whether “more intelligence” becomes a self-justifying ambition whose conditions and consequences escape meaningful negotiation.
When greater capability becomes the dominant measure of progress, who decides whether its costs remain acceptable—and to whom?.
In his March 10, 2026 essay, “AI Is a 5-Layer Cake,” NVIDIA chief executive Jensen Huang describes an industrial stack: energy, chips, infrastructure, models, and applications. Energy powers computation; chips perform it; infrastructure brings together land, power delivery, cooling, construction, and networking. Models operate on that foundation, with applications at the top. Huang presents the layers as mutually reinforcing components of an industrial buildout. This is his account of AI production, not an endorsement of my extractivism critique. [web:74]
Read alongside mineral trajectories, the stack makes apparently immaterial intelligence materially legible. Chips connect computation to mineral supply chains and manufacturing. Infrastructure requires construction, maintenance, and labor. Models bring questions of data, knowledge, permission, and paid work into view. Applications organize access and participation, sometimes creating further opportunities for commercial appropriation. These relationships cross Huang’s layers rather than forming a separate layer of their own. [web:27][web:32][web:74]
My concern is how these mineral and technological infrastructures are governed, and how authority, benefits, and burdens are distributed among the people and communities connected through them. If demand at the application layer justifies expansion beneath it, who can contest the terms? Which communities influence infrastructure decisions? What authority do workers and knowledge contributors retain? Who captures commercial value, and who remains responsible for the system’s material life after replacement?
I tentatively ask whether “AI Obesity” could name an insatiable orientation toward capability, scale, deployment, and computational capacity. It is an exploratory extension of Extractive Obesity, not a finding that all AI development is extractivist. Legitimate research and useful applications must be distinguished from expansion organized through unequal appropriation, concentrated control, or displaced harm. Greater capacity alone establishes neither exploitation nor overshoot.
Two kinds of limit matter. Material and social limits concern identifiable conditions of resources, infrastructure, labor, and exposure. Possible governance limits concern whether institutions and affected populations can evaluate, direct, or constrain deployment. Demonstrating overshoot would require case-specific evidence, not the assumption that rapid growth has already crossed every relevant boundary.
Artificial superintelligence remains a speculative future possibility, not an inevitable destination; pursuing it would not, by itself, establish either overshoot or catastrophe.
The consequential question arises before that hypothetical future. Can technological ambition remain answerable to people whose resources, work, and participation sustain it? Extractive Obesity would describe an appetite that treats expansion as its own justification. Whether that appetite produces Overshoot depends on the demands involved, the limits encountered, and the capacity to change course.
Changing the relationships
A socially beneficial application does not automatically have equitable governance. A system can improve education or safety while leaving ownership, exposure, and decision-making power unchanged.
Te Hiku Media offers a more relevant point of comparison because it explicitly centers Māori data sovereignty. In a 2022 statement, Gianna Leoni described tikanga as foundational to its media and technical work and warned against sharing data and stories without attention to their terms of use. Its Kaitiakitanga License development repository also requires permission to use or adapt the license and cautions that it is unlikely to apply outside an Indigenous context. [web:61][web:71]
These statements are not proof that every decision is equitable. They demonstrate a different starting question: who has authority to determine how knowledge and resources circulate? That is not equivalent to adding an ethical use case after resources have already been appropriated.
The lesson is not to transplant an Indigenous governance arrangement into every technology project. It is to evaluate alternatives through actual authority: who may authorize use, refuse it, negotiate benefits, and demand accountability. Better technical performance matters, but it cannot substitute for those relationships.
Design’s responsibility
Design participates in creating the conditions under examination. Interfaces organize consent; product strategies determine what is measured; engagement goals influence participation; hardware decisions affect repair and replacement. Crawford and Joler’s Anatomy of an AI System makes this interconnectedness visible by tracing labor, data, and planetary resources through a consumer device. [web:27]
We design technological systems, but those systems also shape how we participate and what we come to regard as normal. The point is not that interfaces determine behavior. It is that repeated design decisions can make particular forms of consumption and appropriation routine.
Designers can investigate those relationships and help develop alternatives: reducing unnecessary collection, making refusal usable, examining procurement conditions, extending product life, and involving affected people in consequential decisions. These are directions for inquiry and institutional change, not a claim that design alone can redistribute political-economic power.
Contracts, labor representation, regulation, public infrastructure, and community authority matter. A clearer interface cannot compensate for the absence of enforceable rights.
Beyond extractivist arrangements
The future is not in the end of extraction. It lies in changing the arrangements through which resources are obtained, value is accumulated, and consequences are distributed.
Extractive 2.0 names my interpretation of a new phase in that struggle. Digital systems expand the reach of appropriation into ongoing participation while remaining attached to material trajectories and labor. The enduring logic is concentrated control and displaced responsibility. The transformation is the capacity to make ordinary activity continually productive for systems whose direction participants may have little power to influence.
The expanding pursuit of AI capability sharpens this question; it does not settle it. The test of another future is not whether it uses more advanced technology.
It is whether people who provide resources, perform work, and absorb consequences gain meaningful authority over the relationships connecting them.
Source notes
The linked source markers lead to the original publications. Dates below refer to publication dates where verified; “n.d.” indicates that no publication date was established.
1. Federal Foreign Office, Political Archive. (n.d.). “General Act of the Berlin West Africa Conference, 26 February 1885.” Federal Foreign Office, Germany. Institutional historical account accompanying the treaty material; the date in the title identifies the historical act, not the webpage’s publication date. [web:52]
2. International Telecommunication Union and United Nations Institute for Training and Research. (2024, March 20). The Global E-waste Monitor 2024. ITU/UNITAR. Official publication page. [web:42]
3. International Telecommunication Union and United Nations Institute for Training and Research. (2024, March 20). “Electronic waste rising five times faster than documented e-waste recycling: UN.” Joint report-launch release, published by ITU. Supports the manuscript’s transboundary-movement statement as well as its global waste figures. [web:43]
4. Couldry, N., and Mejias, U. A. (2019, June 30). “Making data colonialism liveable: how might data’s social order be regulated?” Internet Policy Review, 8(2). DOI: 10.14763/2019.2.1411. [web:32]
5. Perrigo, B. (2023, January 18). “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic.” TIME. [web:25]
6. Crawford, K., and Joler, V. (2018). Anatomy of an AI System: The Amazon Echo as an anatomical map of human labor, data and planetary resources. Web essay and visualization, published at Anatomy of an AI System. The source displays the year but not a day-and-month publication date. [web:27]
7. Huang, J. (2026, March 10). “AI Is a 5-Layer Cake.” NVIDIA Blog. [web:74]
8. Te Hiku Media. (2022, May 11). “Māori data sovereignty core focus at Te Hiku Media.” Te Hiku Media. Organizational account reporting Gianna Leoni’s statements; no individual written byline is displayed. [web:61]
9. Te Hiku Media. (n.d.). Kaitiakitanga-License: Repository for the development of the Kaitiakitanga License. GitHub repository, README. Undated, evolving organizational material; consulted October 29, 2026. [web:71]
