While AI is radically altering our economic and societal systems, its impact on the environment is becoming more prominent. In this paper, we will explore the latest findings regarding both direct environmental effects – such as energy use in AI training and inference, emissions from data centers, and hardware manufacturing/waste disposal – and indirect effects related to AI-driven technologies that can either mitigate or relocate emissions.
According to our study, training the model requires a considerable amount of power consumption and production of many pollutants. As was the case in the era of large-scale models and data centers when there were huge demands for materials and waste production.
Artificial intelligence-based applications could contribute positively depending on how they are designed and utilized. In particular, artificial intelligence could increase transportation, agricultural, and industrial processes' energy efficiency. Moreover, artificial intelligence-based systems could be used to monitor environmental compliance by individuals.
Therefore, artificial intelligence's effect on the environment may not necessarily be negative. Its impact is dependent on design and system implementations and policies.
We think a three part framework is a way to look at and improve AI sustainability.
This framework has three parts:
- We need to measure and be transparent about how AI affects the environment. We need to use the same way to measure the impact of computers, energy and materials.
- We need to design AI systems to be more efficient. This means making the models and hardware work together making the models smaller and scheduling the work to use less energy.
- We need to have policies and make sure we use things in a circular way. This means having rules, standards for buying equipment and ways to reuse and recycle AI hardware.
If we use this framework we can make sure that AI innovation is good for the environment and helps us meet our climate goals.
Some important things we need to do include:
- Companies must report how energy they use to train AI models and how much they pollute
- We should give rewards to data centers that use carbon
- We need to invest in ways to reuse and recycle AI hardware
- We need to support research into AI systems that use computer power
We also need to think about fairness. Countries like India need to be able to use AI without hurting the environment so they need to invest in the right infrastructure.
- Energy and Policy Considerations for Deep Learning in NLP Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and policy considerations for deep learning in NLP. Proceedings of ACL.
- Carbon Emissions and Large Neural Network Training Patterson, D., et al. (2021). Carbon emissions and large neural network training. arXiv.
- Green Algorithms: Quantifying the Carbon Footprint of Computation Lannelongue, L., Grealey, J., & Inouye, M. (2021). Green Algorithms. SoftwareX.
- Towards Reproducible and Efficient Deep Learning Henderson, P., et al. (2020). Towards reproducible and efficient deep learning. NeurIPS Workshop.
- Model Efficiency and Hardware Co-design Hsu, C. H., et al. (2022). Model efficiency and hardware co-design. IEEE Transactions on Computers.
- On the Opportunities and Risks of Foundation Models Bommasani, R., et al. (2021). On the opportunities and risks of foundation models. Stanford CRFM.
- ICT for Sustainability: An Emerging Research Field Hilty, L. M., & Aebischer, B. (2020). ICT and environmental sustainability. Environmental Impact Assessment Review.
- Combining Machine Learning with Life Cycle Assessment Wang, X., et al. (2024). Combining machine learning with life cycle assessment. Int. Journal of LCA.
- Circularity in AI Hardware Supply Chains González, J., et al. (2023). Circularity in AI hardware supply chains. Resources, Conservation & Recycling.
- International Energy Agency (IEA Reports) International Energy Agency. (2023). Data Centres and Data Transmission Networks.
- Intergovernmental Panel on Climate Change IPCC. (2023). Sixth Assessment Report.
- United Nations Environment Programme UNEP. (2022). Sustainability and Digital Transformation Reports.
- World Economic Forum World Economic Forum. (2022). Green AI: How to Make AI Sustainable.
