Towards Sustainable AI: Mitigating Carbon Impact Through Compact Models

Abstract

As AI technology continues to advance rapidly, it is essential to address the environmental concerns associated with the increasing carbon emissions and their contribution to global warming. The expanding AI industry requires significant computing power, making it a potential major contributor to carbon emissions in the future. Unfortunately, our current understanding of AI models is very limited. We conducted a comprehensive analysis involving 12 distinct AI models, encompassing object detection, translation, and text-to-image generation tasks. Our findings revealed that smaller AI models can achieve equal or even better results compared to larger models while offering a significant reduction of carbon emissions. This highlights the potential for environmental savings by prioritizing smaller models. These findings underscore the importance of considering the environmental impact of AI models and encourage the adoption of strategies such as using smaller models and optimizing workload schedules to reduce carbon emissions. By prioritizing sustainability in AI development and deployment, we can work towards a greener and more sustainable future.