Current Research Themes and Future Research Needs on Making AI's Energy Consumption Efficient: A Review

While Artificial Intelligence (AI) offers significant benefits, it also poses challenges to environmental sustainability due to increased energy consumption and resulting carbon emissions. Consequently, optimizing AI's energy efficiency has become a critical priority in the scientific community. This article aims to explore current research on enhancing AI's energy efficiency by reviewing 32 relevant literatures. The current themes include the general relevant topic, the impact of AI on the environment, the tools, and the method to reduce carbon emissions. Key areas for future research include comprehensive carbon emission calculations for all hardware configurations, standardizing methodologies for assessing carbon footprints, evaluating carbon emissions across different AI algorithms, developing energy-efficient models for AI systems amidst increasing device penetration, establishing a taxonomy for measuring AI infrastructure energy efficiency applicable across diverse urban settings, and improving energy efficiency in accelerator components such as GPUs, CPUs, and FPGAs.