The recent explosion of AI has given rise to the discipline of AI ethics—the study of ethical and sociological challenges confronting developers, producers, customers, citizens, policymakers, and civil society groups. 

The first wave of AI ethics was concerned with what AI might do, which equated to the ethics of fantastical scenarios like robot revolt. The second wave of AI ethics addressed practical concerns with machine learning (ML) techniques. Finally, it is time to bring in the third wave of AI ethics that directly confronts our day's environmental calamity. It actively tries to engage academics, policymakers, AI developers, and the general public in the environmental implications of AI.

Sustainable development must be the focal point of this third wave. Although there is a rising effort to focus AI utilization on the "positive" end, it is time to go beyond that and address the sustainability of creating and utilizing AI systems in and of themselves. For example, training a single deep learning, natural language processing (NLP) model on a single GPU can result in about 600,000 lb of carbon dioxide emissions, according to a well-known study by Strubell et al. Compared to typical consumption, this amounts to about the same amount of carbon dioxide emissions created over five cars. 

According to the MIT research, Google's AlphaGo Zero’s CO2 emission was equivalent to 1000 hours of air travel or the carbon impact of 23 American homes. So one has to wonder if the emissions from algorithms that can play games (or perform other menial duties) are worth the cost at a time when the world must devote itself to lowering carbon emissions. Furthermore, AI is a technology used in more than just industry or healthcare. It promises to be just as commonplace as smartphones or the internet. We cannot afford to disregard the environmental consequences associated with this technology. 

Research contribution

German researchers propose the following definition of sustainable artificial intelligence: Sustainable AI is an initiative to promote shifts across the whole AI product lifecycle (i.e. creativity, training, fine-tuning, deployment, and governance) that increase ecological integrity and social justice. Sustainable AI is concerned with more than just AI programmes; it also looks at AI's larger socio-technical context. 

In addition, the researchers have defined two subsets of Sustainable AI: AI in the service of sustainability. She suggests redefining sustainable artificial intelligence to centre on the contradictions between AI innovation and equitable resource distribution; inter- and intra-generational justice; and the environment, society, and economy.

Conclusion

According to Aimee van Wynsberghe, "sustainable AI" has two branches: AI for sustainability and sustainability of AI. While the latter is a hidden aspect of evolution, the former has recently attracted much attention.

Furthermore, the researcher claimed that sustainable development, along with three related tensions between AI innovation and just resource allocation, inter- and intra-generational justice, and tensions between the environment, society, and economy, form the basis of the definition of sustainable AI. The three pillars of sustainability—social, economic, and environmental—and sustainable AI—are not intended to be discussed in the suggested research article. Instead, the purpose of the study is to remind the reader, policymakers, AI ethicists, and AI developers that there are environmental consequences associated with AI.