AI’s extractive core: Data as digital frontier of resource colonialism

A new study published in AI & Society challenges conventional views of artificial intelligence (AI) by asserting that data is not immaterial, but a fully material category, one that parallels the extraction of resources, labor, and energy. The research paper, titled “Is Data Material? Toward an Environmental Sociology of AI”, provides a theoretical foundation for environmental sociology to engage critically with AI, using the dual concepts of abstraction and extraction to redefine the materiality of data within digital capitalism.

The author critiques the prevailing disconnect in academic and public discourse that treats data as disembodied or virtual, arguing instead that it is embedded in physical infrastructures and violent socio-political processes that warrant urgent scrutiny.

Why has AI’s material impact been overlooked?

The study identifies two main barriers that obscure AI’s materiality: the narrow focus on data extractivism and the prevailing tendency to consider data as immaterial. While critical AI literature often critiques the extractive nature of data, Maximilian Pieper, the author, highlights a gap in understanding how the data extraction process is itself materially grounded.

He points out that this oversight stems in part from the intellectual legacy of Big Data studies, which have shaped much of contemporary AI discourse. With AI now popularly understood as an umbrella term for advanced data processing using neural networks, such as those behind ChatGPT or DALL·E, the nuanced material and infrastructural dependencies of these systems are often flattened or ignored. In critical literature, data is frequently collapsed into other extractive categories such as resource mining, energy consumption, or outsourced labor, further distancing it from its own material identity.

According to Pieper, this approach reinforces a problematic binary between the “digital” and the “analog” worlds. He argues that data should be analyzed not as a derivative of other materials but as a distinct and fully material form shaped by specific processes of abstraction and social relations.

What makes data extraction material and political?

Pieper’s argument is fundamentally grounded in the Marxist-influenced idea of “real abstraction.” He reinterprets data through this lens, proposing that its abstraction is not merely conceptual but enacted through social processes. Much like the way commodities are imbued with value through exchange, data is materialized through its integration into technical systems that reduce complex realities to functionally useful elements.

This perspective positions data alongside energy, labor, and natural resources as functional simplifications of a richer, more complex world. Whether it’s a lithium mine in Serbia, the energy needs of server farms, or gig workers labeling images for machine learning, these are not parallel processes but deeply entwined ones. Data emerges from these entanglements as both product and method of abstraction.

Pieper identifies this reduction as a form of violence, not metaphorically, but in its direct consequences for people and ecosystems. The extraction of data disembeds human actions and natural phenomena from their contexts to transform them into machine-readable inputs, often stripping them of meaning and agency. For instance, screen time on social media is not a passive act; it is labor, surveillance, and monetization simultaneously.

Importantly, the study underscores how abstraction allows technologies to function without acknowledging the sociopolitical and ecological costs of that functionality. Technologies like predictive policing or algorithmic content recommendation systems operate through models that simplify and reify human behavior, often reinforcing systemic biases and inequalities. These systems function precisely because they ignore complexity - a feature, not a bug.

Can Environmental Sociology Address AI’s Extractivism?

The author calls for environmental sociology to step into this analytical void. The discipline, he argues, is uniquely equipped to investigate the socio-ecological consequences of technological abstraction. Rather than focusing merely on more ethical or efficient AI systems, environmental sociology can interrogate the underlying social structures that make AI “function” in the first place.

He critiques mainstream responses that reduce the issue to quantifying CO₂ emissions or energy costs. While such metrics are important, they miss the broader question: how are human lives and ecosystems reshaped to serve the technical requirements of AI? He suggests that counting kilowatt-hours or gigabytes is insufficient without understanding the political economy and social dynamics of abstraction that render such data extractable in the first place.

The author's approach is at odds with relational ontology frameworks, such as actor-network theory or posthumanism, which emphasize the material agency of data and technologies but often downplay the role of conflict and inequality. Instead, he proposes that abstraction be seen as a form of commodification deeply rooted in colonial logics of dispossession and value extraction.

The study establishes a direct link between modern AI extractivism and colonial practices of statistical enumeration, land ownership, and epistemic control. Pieper notes that many of the corporations extracting and monetizing data today are headquartered in the Global North, continuing a pattern of resource and labor appropriation that defined earlier colonial regimes. However, he cautions that data’s extraction dynamics differ from classical colonial forms and should not be oversimplified. For example, most large-scale data generation and consumption still occurs within the Global North, complicating traditional North–South dichotomies.

The study also engages with ecofeminist and decolonial critiques by highlighting how the “more-than-functional” aspects of life, such as social reproduction, care work, and ecological stewardship, are systematically devalued or excluded from commodification.