Scope
In (Kaack et al, 2022) provides a systematic framework for analysing the relationship between ML and climate change mitigation. The authors classify the impact of ML on greenhouse gas GHG emissions into three categories: (A) compute-related emissions from hardware and energy use, (B) emissions from the immediate applications of ML, and (C) system-level impacts, such as societal or rebound effects. The paper emphasises the need for holistic assessment methodologies and policies that align AI development with climate mitigation strategies.
Summary
In this work several key insights are provided. First, there is the issue of rebound effects, where improving the efficiency of ML models may counterintuitively increase overall consumption, leading to higher emissions—a challenge that must be carefully managed. Second, system-level impacts of AI can have more significant consequences than direct compute-related emissions. For instance, while optimising transportation systems through AI can reduce emissions, increased consumption driven by advertising algorithms may have the opposite effect. Finally, path dependency and lock-in pose another challenge, as the widespread adoption of ML in certain industries may cumulate emissions-intensive technologies, hindering the transition to more sustainable alternatives.
Relevance for EXIGENCE
This paper is highly relevant to the EXIGENCE project for energy ecosystem, energy optimisation and standardisation; as it highlights critical insights into the environmental impact of AI and ML, which is a key concern for EXIGENCE’s focus on energy-efficient technologies. Concepts like rebound effects and system-level impacts discussed in the paper mirror similar concerns within EXIGENCE, where improving the efficiency of ML models might inadvertently increase overall consumption and emissions. The paper’s emphasis on the need for energy-efficient AI and the development of sustainable computational practices directly supports EXIGENCE’s broader mission of optimizing energy usage in technology-driven industries.
- H. Kaack, P. L. Donti, E. Strubell, G. Kamiya, F. Creutzig, and D. Rolnick, “Aligning artificial intelligence with climate change mitigation,” Nature Climate Change, vol. 12, no. 6, pp. 518–527, Jun. 2022.
