The Environmental Cost of Engineering Machine Learning-Enabled Systems: A Mapping Study

What is it about?

The study explores the environmental impacts of integrating machine learning (ML) into software systems, a process increasingly common in both public and industrial sectors. This integration has raised concerns about sustainability, particularly due to the substantial energy consumption required for the ongoing training, testing, and retraining of ML systems. ML-Enabled Systems (MLES), utilize large-scale data processing and complex algorithms, often needing powerful hardware like GPUs (Graphics Processing Units) which are energy-intensive. The frequent and intensive computation not only increases electricity use but also contributes to higher carbon emissions. This paper addresses these issues by presenting a comprehensive review and a systematic mapping study of the current research focused on sustainable AI. It examines techniques, tools, and learned lessons that could help assess and improve the environmental sustainability of these systems, particularly within the MLOps (Machine Learning Operations) framework. MLOps is a set of practices that aim to automate and optimize the ML lifecycle in production environments. The study highlights the growing need to mitigate the environmental impact associated with the development and operational phases of AI-powered software, emphasizing the importance of incorporating sustainability into the engineering and maintenance of machine learning systems.

Featured Image

Photo by Zbynek Burival on Unsplash

Photo by Zbynek Burival on Unsplash

Why is it important?

It addresses the significant environmental impact of Machine Learning-Enabled Systems, particularly the growing carbon footprint from the frequent training and operation of these systems. It offers a comprehensive examination of the techniques and tools available to assess and promote sustainability in MLOps practices, which is crucial for the responsible development and deployment of AI technologies. The study not only offers an overview in current research but also gives practical insights that could help in developing more sustainable AI applications.

Perspectives