The Environmental Benefits and Harms of AI: Preliminary Results from a Review Study

In recent years, the environmental impact of Artificial Intelligence (AI) has garnered significant attention, leading to the emergence of Green AI research. Green AI focuses on mitigating the carbon footprint of AI systems, driven by the recognition that the computational resources required for AI operations are substantial and growing. The current body of research primarily addresses the operational phase of AI systems. These studies emphasize the importance of monitoring energy consumption, optimizing model parameters for sustainability, and developing tools to measure and reduce the carbon emissions associated with AI training and inference processes. Despite the extensive research conducted on the environmental impact of AI operations, the effects of AI hardware production, logistics, deployment within industry, and end-of-life management remain overlooked within the Green AI literature. Therefore, we seek to broaden the definition of Green AI to these other aspects and rely on a tertiary literature review to create an evidence base to support this. We ask, 1) What are the key environmental sustainability benefits and harms associated with AI production, logistics and hardware end-of-life? 2) What are the key indirect environmental sustainability impacts associated with AI deployment across various sectors and contexts?

In this tertiary review we aim to gather and synthesize the literature in order to develop a robust understanding of AI’s environmental impact and foster the development of effective mitigation strategies where necessary. We include in our analysis downstream societal impacts that may follow environmental impacts. Our review includes existing peer-reviewed secondary studies (reviews) and grey literature and is carried out in accordance with the PRISMA guidelines for systematic reviews. In this presentation, we provide preliminary findings that illustrate the sustainability risks and challenges associated with AI hardware production, logistics and end-of-life, and indirect environmental benefits and harms that stem from AI deployment in various industrial contexts.