Environmental Impacts of AI

Summary

The global proliferation of artificial intelligence (AI) is accompanied by significant debate regarding its net environmental consequences. AI is lauded as a potential panacea for climate change—enabling smart grids, maximizing resource allocations, reducing operational waste, and accelerating materials discovery for carbon reduction. Yet AI’s potential climate mitigations sit uneasily against AI’s extractive and consumptive nature. AI is reliant on environmentally fraught critical materials for hardware, toxic manufacturing processes for chips and servers, and energy-and water-intensive data centers for data processing, all while contributing to the planet’s fastest-growing waste stream: e-waste. Additionally, AI applications are being used to accelerate oil and gas extraction and consumption, ultimately speeding up global carbon emissions production. Materialist approaches to the study of science and technology provide important frameworks for understanding the complex ecologies of AI—from macro-scale analyses of the political economy of AI to micro-scale analyses of the situated impacts of AI’s hardware, data centers, energy infrastructures, water consumption, carbon emissions, land disruptions, noise and light pollution, and community frictions that arise from AI’s material footprint. Scholarship on AI’s environmental impacts can be organized into four categories, or scopes, which follow established environmental reporting standards: Scope 1, direct impacts from AI’s data centers; Scope 2, indirect impacts from energy infrastructures; Scope 3, supply chain externalities; and the emerging field of Scope 4, afforded impacts, encompassing both enabled mitigations and enabled harms of AI applications. While Scopes 1–3 are increasingly documented in scholarship, Scope 4 remains under-analyzed despite its potential to dominate AI’s long-term ecological consequences. Critical Science, Technology, and Society (STS) perspectives on AI’s environmental impacts have highlighted the extractive elemental, infrastructural, and resource value chains that AI relies on. This scholarship has done much to reveal AI’s hidden material dependencies. Yet future research opportunities exist to deepen analysis of power relations, systems of dependency, local and global inequities, and how the political economy of planetary computation has expedited the concentration of natural and human resources into the hands of few corporations at the expense of the global majority.