Is data material? Toward an environmental sociology of AI
- Critical social science studies on AI often focus on its data extraction. However, this process is rarely understood as material. When scholars do look at the material extraction on which AI technologies rely, they tend to dissolve the category of data into established forms of extraction—for example of resources, energy, or labor time. Notwithstanding that they constitute a crucial part of AI’s materiality, dissolving the category of data into seemingly more material ones seems unnecessarily limiting and reproduces the illusory divide between an immaterial digital and a material ‘analog’ realm. I argue that the concept of a dual process of abstraction and extraction, commonly evoked in literature on data extraction, can help to conceptualize the materiality of extraction as a process by which reality is narrowed to a set of functional properties, while disregarding everything else. In the case of data, this process has unique dynamics that make it distinct from, yet equally materialCritical social science studies on AI often focus on its data extraction. However, this process is rarely understood as material. When scholars do look at the material extraction on which AI technologies rely, they tend to dissolve the category of data into established forms of extraction—for example of resources, energy, or labor time. Notwithstanding that they constitute a crucial part of AI’s materiality, dissolving the category of data into seemingly more material ones seems unnecessarily limiting and reproduces the illusory divide between an immaterial digital and a material ‘analog’ realm. I argue that the concept of a dual process of abstraction and extraction, commonly evoked in literature on data extraction, can help to conceptualize the materiality of extraction as a process by which reality is narrowed to a set of functional properties, while disregarding everything else. In the case of data, this process has unique dynamics that make it distinct from, yet equally material as, resource, energy, or labor extraction. Connected to the Marxist concept of ‘real abstraction’, such approach is sensitive to power relations and helps to critically investigate depoliticized notions of technological functionality. The materiality of AI does not exhaust itself in the quantities of kilograms of raw material, megajoules of electricity, or labor hours. An environmental sociology of AI would instead focus on the socio-ecological processes through which people and the planet are pressed into these functional abstractions in the first place.…
