Green AI: Systematic Review and Guidelines for Sustainability in Artificial Intelligence
Authors: Hareem Nisar*, Austin Tapp*, and Marius George Linguraru Institute: Sheikh Zayed Institute for Pediatric Surgical Innovation, Children's National Hospital, Washington, DC, USA
Abstract: Artificial intelligence (AI) is now a planetary-scale socio-technical infrastructure whose energy demands are rising faster than our capacity to measure or mitigate them. Large-scale commercial applications like Claude and ChatGPT have made AI widely accessible, however, generative AI models are also avid electricity consumers and carbon emission contributors. Naturally, the question of how we might promote energy-friendly ‘green’ AI arises. In this systematic review, we synthesize the last seven years (2017–2024) of peer-reviewed work describing how AI’s energy use and emissions are measured, how substantial these emissions are across the AI lifecycle, and what techniques can reliably reduce AI-related carbon emissions. We identified five distinct themes that emerge in the context of green AI and have the potential to reduce energy consumption. The observations and analysis of our review suggest a lack of standardization in measuring and reporting energy cost and carbon emissions associated with the AI lifecycle. To address this dearth in reporting standards, we propose eight review-driven, actionable guidelines for researchers, industry, and policymakers to promote environmentally sustainable and green AI as a proactive property of the AI lifecycle.
Note: This review article is currently under consideration at a journal. It is shared here as a preprint for scholarly communication only. No permission is granted to reuse, redistribute, or create derivative works from this material without the authors’ prior written consent.
*(equal contribution)