top of page

Environmental Impacts

Ethical actions require mindful consideration and transparency. While not necessarily solutions in and of themselves, these are essential building blocks for sustainable practices. A key area for deliberation concerns the environmental impacts of AI’s increasing usage. While often marginal compared to other forms of energy consumption, the energy needed to train and run machine-learning systems continues to steadily increase [49].


One study in particular demonstrates how AI’s energy consumption has “doubled roughly every 3.4 months—increasing 300,000 times between 2012 and 2018” [50]. Part of the reason for this exponential growth is the development of more advanced systems, such as highly accurate natural language processing systems (NLPs) that require massive “computational resources that necessitate similarly substantial energy consumption” [51]. This significant energy usage is costly in both a financial sense and an environmental one “due to the carbon footprint required” to power these systems [52].


Numerous studies have drawn parallels between the demands of such systems and other forms of human activity. Some of these statistics can appear quite alarming, particularly given the rapid spread of AI. However, these concerns should be tempered by a few important observations.


Firstly, the activism of individuals and certain organizations has already garnered a great deal of traction towards securing a sustainable future for AI. Sasha Luccioni has led an effort to create a website where companies can calculate the carbon footprint of their machine learning efforts. As she explains, “Hopefully this will go toward full transparency [...] So that people will include in the footnotes ‘we emitted X tons of carbon, which we offset’” [53]. Transparency endorses accountability, which in turn promotes tempered progress.


While headway certainly has been achieved, and many large corporations already claim carbon neutrality or are openly working towards it, there is still substantial room for growth. Cloud providers, for example, generally do not “disclose the overall energy demands of machine-learning systems” [54]. Although they may not be entirely open about the impacts of their efforts, decreases in costs do function as a non-negligible motivation for companies to reduce their energy consumption.


One immensely encouraging fact is that AI can be an excellent tool for increasing efficiency and reducing waste. In fact, “AI-enabled use cases have helped organizations reduce GHG [Greenhouse Gas] emissions by 13% and improve power efficiency by 11% in the last two years” [55]. The potential for AI to help reduce energy inefficiency in a number of sectors is exciting, to say the least. But we must continue to hold ourselves accountable, both in our use of AI and in our allocation of resources and attention. This is an issue that will not simply be resolved through the implementation of AI. Rather, personal and social accountability must always come first.


bottom of page