Practical Evaluation of Machine Learning Efficiency Requires Model Life Cycle Assessment

Jared Fernandez, Clara Na, Yonatan Bisk, Constantine Samaras, Emma Strubell

Keywords: position paper, life cycle assessment, efficiency, energy efficiency, socioeconomic impact, environmental impact, holistic evaluation

TL;DR: We propose life cycle assessment (LCA) for LMs/ML models, introducing examples of concrete empirical benefits of adopting an LCA framework for efficiency research, and calling for community-wide participation in welcoming future work in this vein

Abstract: The growing scale of language models entails growing resource requirements and envrionmental impacts. For these systems to have a positive impact on society, it is necessary to thoughtfully weigh the societal and environmental benefits and costs, within the context of a complex model life cycle and many potential measures of impact. In this position paper, we argue the need for $\textbf{holistic life cycle assessment of language models}$ across the development and deployment pipeline to properly account for required resources and downstream impact.

Submission Number: 24