Abstract
As artificial intelligence continues to play a growing role in everyday life, it is important to examine not only how these systems are built, but also the infrastructure and environmental impacts that support them. As an engineering student, I have explored this issue from both a technical and sociotechnical perspective. My technical project develops a systems-level framework for measuring the environmental impact of AI, with an emphasis on analyzing energy use, water consumption, and scaling effects across large-scale infrastructure. My STS research examines Google’s AI sustainability claims, focusing on data center expansion in Northern Virginia and using Actor-Network Theory (ANT) to analyze how metrics, reports, and policy actors shape environmental narratives and governance decisions. Together, these projects show that how environmental impact is measured directly influences how AI systems are evaluated, expanded, and justified. For engineering, this highlights the importance of not only improving technical performance, but also critically examining the measurement systems that guide decision-making and shape real-world outcomes.
The technical portion of my project focuses on developing a systems-level framework for measuring the environmental impact of artificial intelligence. Current approaches often rely on per-query or per-task efficiency metrics, which make individual uses appear minimal while overlooking total system demand. To address this, my framework evaluates AI systems at scale by incorporating total energy consumption, water usage, and lifecycle impacts across training, deployment, and continued operation. It also accounts for how increased efficiency can drive greater adoption, leading to higher overall resource use rather than reduction. The results show that efficiency improvements do not necessarily translate to lower environmental impact when systems continue to expand. This is significant because it highlights the need for more comprehensive measurement methods that better reflect real-world infrastructure demands and can inform more responsible engineering and policy decisions.
With this in mind, my STS project complements my technical portion by diving into one of the major AI developers, Google. Specifically, my project looks into Google’s claims of AI sustainability. Using the ANT framework, I analyze how both human actors, such as engineers, policymakers, and regulators, and nonhuman actors, such as efficiency metrics, environmental reports, and infrastructure systems, interact to shape these claims. The project focuses on data center expansion in Northern Virginia, a region containing one of the highest concentrations of data centers, making the environmental and infrastructural impacts more visible. Rather than simply evaluating whether Google’s claims are accurate, I examine how these claims are constructed through metrics, reports, and policy processes, and how they influence decision-making around energy use and expansion. This matters because it shows that AI sustainability is not just a technical issue, but also a question of governance and accountability, where the way impact is measured can determine environmental outcomes.
Together, these projects highlight how technical measurement and social interpretation are closely connected in shaping the impact of artificial intelligence. While my technical work focuses on improving how environmental impact is measured at a systems level, my STS research shows how existing metrics are used in practice to frame sustainability and guide decision-making. This connection reveals that measurement is not neutral, as the way engineers report efficiency can influence infrastructure expansion, policy decisions, and public understanding. For engineers, this emphasizes the importance of considering not only technical performance, but also how metrics affect broader social and environmental outcomes. An STS perspective reinforces that engineering decisions exist within larger networks of stakeholders, institutions, and material systems. Ultimately, combining technical and socio-technical approaches supports more transparent development of AI systems by aligning engineering practices with long-term environmental impacts.
