Artificial intelligence (AI) is increasingly promoted as a lever for the “green–digital” transition, yet its sustainability remains contested. This paper pursues a two-part analysis: AI for sustainability (uses of AI to achieve domain goals such as grid optimization or climate adaptation) and the sustainability of AI (the environmental and social implications of AI’s own life cycle). The paper argues that domain benefits do not, by themselves, establish sustainability. Drawing on the distinction between thin and thick sustainability, it demonstrates that narrow indicators or alignment on the sustainable development goals (SDGs) can obscure rebound effects, material throughput, and distributional asymmetries. The discussion addresses AI’s energy and environmental costs such as high compute training, water- and carbon-intensive data centers, hardware manufacture and disposal. It situates these within patterns of ecologically unequal exchange. Furthermore, three ‘materialities’ of AI are also explained in more detail: physical (critical minerals and energy), informational (data practices and infrastructures), and social (labor and governance), to show how benefits and burdens are unevenly distributed. The paper concludes that AI can be regarded as sustainable only when it both materially advances sustainability goals and meets robust life-cycle criteria. Where these conditions are not met, smaller models or non-use of AI should be considered.
Eichentopf et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: