Position: Carbon Footprint Reporting Should Be Routine in Machine Learning Research

Guan-Ming Chiu

TL;DR: This position paper argues that ML papers should report carbon footprints as standard scientific practice.

Abstract: We argue that the machine learning community should adopt standardized carbon footprint reporting as part of routine scientific practice. Training large models can emit hundreds of tons of CO$_2$, yet environmental costs remain largely invisible in publications, leaving efficiency claims scientifically incomplete and reproducibility undermined when identical experiments in different locations yield vastly different carbon footprints. We put forth reporting guidelines comprising five standardized metrics, practical measurement tools, and integration with community benchmarks, with a phased three-stage adoption process. We address alternative views, including measurement complexity and barriers for resource-limited researchers. To promote equity, we advocate for dual reporting of energy and carbon, reference-grid normalization, and acceptance of approximate estimates. This paper calls on venues, reviewers, authors, and institutions to establish carbon awareness as a foundational element of responsible ML research.

Lay Summary:

Training modern AI models consumes vast amounts of electricity and can release hundreds of tons of carbon dioxide into the atmosphere, comparable to the lifetime emissions of dozens of cars. Yet when researchers publish AI papers, they almost never report how much energy their experiments used or how much carbon was emitted. This blind spot has real consequences: when a paper claims a new method is "more efficient," it is often unclear what "efficient" actually means, since the same computation in one country can produce 25 times more carbon than in another simply because of local electricity sources.

In this position paper, we argue that the machine learning community should make carbon footprint reporting a routine part of publishing research, in the same way that papers already report hardware details and training time. We propose a practical set of five standardized things to report (energy used, carbon emitted, electricity-grid carbon intensity, data-center efficiency, and computing location), a list of free open-source tools that can measure them automatically, and a gradual three-stage plan for conferences such as ICML to encourage adoption.

We also address common objections, including concerns that measurement is too difficult or that reporting may disadvantage researchers at smaller institutions or in regions with dirtier electricity grids. To keep the system fair, we recommend reporting energy alongside carbon, normalizing comparisons to a common reference grid, and accepting approximate estimates rather than demanding perfection. Our call is directed at conferences, reviewers, authors, and institutions: making the environmental cost of AI research visible is a necessary step toward responsible and scientifically honest progress.

Link To Code: https://github.com/guan404ming/carbon-footprint-reporting

Primary Area: Social, Ethical, and Environmental Impacts

Keywords: Green AI, Carbon Footprint, Sustainable Computing, Energy Efficiency, Responsible AI

Submission Number: 983