AI and Sustainability
The energy requirements of digital infrastructure are rising continuously; currently, the digital infrastructure worldwide emits twice as much CO2 as civil aviation. The increasing use of AI reinforces this trend, as the training of algorithms in particular is very energy-intensive. At the same time, AI algorithms can make significant contributions to the efficiency of energy use, or be used in data collection to analyze environmental changes. Therefore, from a sustainability perspective, a differentiated view on AI is necessary.
The perspectives on AI & sustainability can be divided into two approaches. On the one hand, perspectives that focus on the potentials of the technology for a transformation towards more sustainability (AI for sustainability) and, on the other hand, perspectives that deal with the (also unintended) impacts of AI applications on sustainability (Sustainability of AI) (cf. van Wynsberghe 2021).
Research on the potentials of AI for sustainability shows a wide range of possibilities in which algorithms can be applied, for example to improve agricultural yields, for reductions in resource consumption such as water or energy, and for organizing transportation. The networking of systems, for example in so-called smart cities, is also a possible application in which AI systems can contribute to increase sustainability. Another strand of research in this field is concerned with research on climate change and the prediction of changes caused by climate change.
Research on the sustainability of AI systems focuses on the impact of the technology on the environment and people. An important starting point is the high energy consumption of machine learning. As Strubell et al. (2019) showed, the computing power used to train NLP systems, for example, generates relevant CO2e emissions. Thus, the training of the NAS system generated about 284t CO2e. In addition to these emissions, there are also effects on the people in the mining areas in the supply chains of the required hardware and the necessary equipment for the AI systems. The analysis of the sustainability of AI systems also includes possible negative effects on certain groups in society. In this context, the analysis of the adoption of discriminatory patterns from training data plays a central role.
- van Wynsberghe, A. (2021): Sustainable AI: AI for sustainability and the sustainability of AI. AI Ethics 1, 213–218.
- Strubell, E.; Ganesh, A.; McCallum, A. (2019): Energy and Policy Considerations for Deep Learning in NLP. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3645–3650, Florence, Italy. Association for Computational Linguistics.