Scope
The article (Nature Editorial, 2022) emphasizes the urgency of limiting global warming to 1.5C, citing the Intergovernmental Panel on Climate Change (IPCC) report which stresses the need for global greenhouse gas emissions to peak by 2025 and to fall by 43% by 2030. The editorial explores several areas where ML can contribute to this goal, such as renewable energy optimization, greenhouse gas emission monitoring, material discovery for batteries, and energy forecasting.
Summary
The editorial highlights contributions from initiatives like Climate Change AI and the Open Catalyst Project, which demonstrate how ML can be leveraged to reduce emissions. However, it also stresses that ML is not a “silver bullet” for climate change mitigation. In fact, ML models themselves can have significant environmental costs due to the carbon footprint associated with training large models. The editorial advocates for holistic assessments of ML’s impact and encourages researchers to report the carbon emissions of their models. It concludes by calling for a shift in focus from accuracy improvements to a more balanced approach that considers both accuracy and energy consumption.
Moreover, ML can play a pivotal role in various sectors by enabling efficient renewable energy distribution, optimising energy storage systems, and discovering new materials for energy conversion. It can also help track deforestation and monitor greenhouse gas emissions through satellite data and remote sensing. However, ML has limitations. For instance, ML is currently used in applications that may worsen climate change, such as in fossil fuel exploration. Furthermore, the environmental cost of training large ML models can be high due to significant energy consumption. Finally, the editorial calls for researchers to report the carbon emissions of their ML models in scientific publications. Initiatives like the NeurIPS workshop ‘Tackling Climate Change with Machine Learning’ are crucial in developing metrics to evaluate the environmental impact of ML methods.
Relevance for EXIGENCE
By emphasising the need for transparency in the reporting of ML’s carbon footprint, the article supports the idea of tracking energy consumption to incentivise low-carbon alternatives. Additionally, it advocates for balancing performance with environmental impact, which aligns well with the goal of encouraging energy-efficient usage of digital resources. Therefore, is highly relevant with the EXIGENCE project, where the focus is to incentivise users to minimise energy consumption and reduce carbon emissions.
662. Editorial, “Achieving net zero emissions with machine learning: the challenge ahead”, Nature Machine Intelligence, vol. 4, no. 8, pp. 661–662, Aug. 2022.
