Towards Sustainability of AI: A Systematic Review of Exisiting Life Cycle Assessment Approaches and Key Environmental Impact Parameters of Artificial IntelligenceDusan Dokic; Florian Groen in't Woud; Wolfgang Maaß
In: Proceedings of the 57th Annual Hawaii International Conference on System Sciences. Hawaii International Conference on System Sciences (HICSS-2024), Digital Innovation for a Sustainable Economy, January 3-6, Waikiki, HI, USA, ScholarSpace, 1/2024.
Most people are aware of the huge benefits that Artificial Intelligence (AI) brings to humanity in terms of sustainable applications (AI for sustainability). Yet, the fewest face the environmental impacts caused by an AI over its complete lifecycle (Sustainability of AI), e.g., the energy consumption, regardless how beneficial its outputs are. This paper presents a systematic literature review on the existing approaches for conducting a Life Cycle Assessment (LCA) on AI applications, alongside the key factors influencing their environmental impact. The study identifies critical environmental impact drivers of an AI over its life cycle, like the energy and resource consumption of hardware devices which provide the needed computing power. It underscores the importance of a holistic LCA approach considering operational and embodied energy use and the lifecycle impacts of data centers and other physical devices required for AI. The results provide critical insights for stakeholders looking to assess and mitigate the environmental impact of AI applications.