GREEN ARTIFICIAL INTELLIGENCE
Lucía Ortiz de Zárate Alcarazo
AI is more than a group of technologies capable of taking decisions as a person would in the same context, employing a range of techniques such as machine learning, deep learning, neuronal networks, etc. It is also a mindset, a way of understanding and exercising power, a mega-machine and an industry based on extractivism (Crawford, 2021). Although these descriptions of AI are both correct, the first is used more frequently than the second. However, to grasp the scope, the ethical-political and, of course the ecosocial implications of AI, the second is more appropriate.
From the point of view of the potential benefits these technologies can generate in areas such as health, education, administration and management of services, etc., and despite the many ethical-political problems enveloping AI, it has managed to find its way into our societies and become an objective to be met and an expectation. Such implementation within society is largely due to the fact that in our technological worldview we associate such ‘new’, ‘digital’ or ‘disruptive’ technologies with ideas that bring to mind progress, development, what is modern, and a clean, environment-friendly future, also known as ‘green’ (Ensmenger, 2018). In this worldview, Green AI plays a lead role as the enabling driver and creator of this present/future. Below, this concept is expanded and some implications are discussed.
1. Green AI
To most people, the relationship between AI and nature is non-existent. This is because, since the end of the 20th century, the ill-named new technologies have been presented as radically opposed to the industrial technologies of the first and second industrial revolutions. Whereas the latter are seen as pollutant and dirty, based on burning fossil fuels such as oil or coal, the former, that include AI, are supposed to be clean and green.
According to this idea, then and now held mainly by economists, technologists, thinkers, academics and politicians, digital technologies and AI are portrayed as practically independent of polluting material infrastructures and exercising minimal ecological impact (Negroponte et al., 1997). This is the gist of all the strategies that, littered with ecological metaphor, associate AI with “the cloud” and centres at which great quantities (“mountains”) of data are presented as “data farms” (Carruth, 2014). However, in recent years, evidence of the negative ecological impact of AI and the technological industry as a whole has brought to light the less attractive side of these new developments. In this regard, forecasts for the technological industry are especially alarming: by 2040 it will be responsible for 14% of greenhouse gas emissions (Belkhir & Elmeligi, 2018). As for AI, in 2018, the centres on European soil alone accounted for 2.7% of the demand for electricity by the entire European Union (EU), a figure that may increase to 3.21% or 6% in 2030, depending on the energy efficiency gained at said centres (Montevecchi et al., 2020).
Against this backdrop, the notion of a green AI has gained relevance among environmentalists (and ecomodernists) who still support the “lightness” of this technology and endorse the use of AI as the solution to many problems, including ecological ones, and among ecologists who are inclined to re-think the current approach to AI and the technological industry as a whole.
2. Environmentalists
Environmentalists generally use the term green AI to refer to neutral energy consumption systems that do not leave a carbon footprint. In this sense, they ignore some of the premises and assumptions that encourage a utopian and unrealistic image of AI. Nevertheless, environmentalists view the relationship between AI and nature as limited and, having set the appropriate coordinates, deem it positive in global terms. From this angle, measures to develop green AI advance in three main directions, nearly all in relation to its “use phase”.
The first measure calls for reducing consumption at the data centres on European soil. The second involves developing “green algorithms” whose energy consumption during training and regular operation is minimal (Schwartz et al., 2020). Lastly, the third entails promoting research focusing on the development of materials that are more recyclable and environment-friendly, with a view to optimising the AI e-waste phase.
This stance, adopted by a majority within the EU and a large part of the scientific and academic community, acknowledges that AI poses some ecological problems, but questions neither the power relations underlying AI, nor the ecological implications of AI beyond the western culture, nor the socioeconomic dynamics in which the development of these technologies are rooted. Therefore, not only does it uphold green AI as the solution to any environmental problems it may cause, but also calls for the use of AI technologies to reduce energy consumption in other sectors, such as optimised water use in the irrigation of public gardens, real-time detection of water leakages, developing models to predict the effects of climate change, etc. (Cowls et al., 2021). From here, green AI adheres to the group of measures designed to make economic, technological and industrial growth compatible with caring for the environment, which gives rise to buzz-words such as ‘AI for Sustainability’, ‘AI for Climate’, or ´AI4Good’.
3. Ecologists
Ecologists, for their part, argue that the environmentalists’ perspective is reductionist and falls short offering partial solutions to a problem that cannot be solved by enhancing the efficiency of AI used in the Western world, but rather by abandoning capitalist rationales for unbridled technological growth and comprehensively regarding global production, manufacturing, consumption and waste generation of AI (Almazán, 2021a). For this group, the relationship between AI, nature and the ecosocial crisis is far narrower, deeper and more troublesome than environmentalists claim. Thus, ecologists advocate studying the phenomenon beyond the “use phase”, which is restricted to rich countries; to input reflections that question the necessity, scope and rate of technological transformation; to challenge preconceived ideas about AI, growth, and their relationships to the myth of progress; and to propose economic, social and political models that are ecologically fair (Almazán, 2021b).
With regard to the life cycle of AI, despite reporting an ecological impact during the use phase, ecologists seek to enlarge their focus to encompass the remaining two phases in the cycle. On the one hand, the initial phase consists in extracting the raw materials and minerals necessary for building AI technologies, as well as their fine-tuning, manufacture and assembly. This phase, usually conducted in countries in the global South, has severe repercussions in the socioecological environment, such as erosion, loss of biodiversity, devastation of surrounding vegetation, water pollution, deforestation, forced labour, etc., all deriving from mining activities and the transportation of components for AI, among others (Dhar, 2021). On the other hand, ecologism also highlights the need to examine the final phase of AI involving the management of e-waste generated at the life-end of the technological devices used by AI, or when these are renovated. On this point, the problems involve the socioecological impact generated by the lack of recycling for these materials and the disposal of e-waste in countries like Ghana and Pakistan (Crawford & Joler, 2020), as well as the over-consumption of technology in the West.
From this baseline, ecologists seek to show that AI, far from being clean and independent of infrastructures, is dirty and dependent on mining, maritime transport, coal and oil. Therefore, the measures to be taken in addressing the ecosocial crisis in relation to AI cannot be limited to proposals for green AI. From this perspective, a green approach to AI cannot be limited to seeking progress in technological efficiency or proposing technological solutions to ecological problems, for which the technology industry itself is largely responsible. Ecologists deem it necessary, as a minimum, to question the growth dynamic in technological production and consumption, to address the green colonialism issue, and to re-define our relationship with the rest of non-human living beings, in order to detach ourselves from a socioeconomic system that has proven, over and over, its incompatibility with life and eco-social justice.
Bibliography:
Almazán, A. (2021a). ¿Verde y digital? Viento Sur, 173.
Almazán, A. (2021b). Técnica y tecnología: Cómo conversar con un tecnolófilo. Taugenit Editorial.
Belkhir, L., & Elmeligi, A. (2018). Assessing ICT global emissions footprint: Trends to 2040 & recommendations. Journal of cleaner production, 177, 448-463. https://doi.org/10.1016/j.jclepro.2017.12.239
Crawford, K. y Joler, V. (2020). Anatomy of an AI system. The Amazon Echo as an anatomical map of human labor, data and planetary resources. Retrieved June 6, 2023, from https://anatomyof.ai/
Crawford, K. (2021). The atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.
Carruth, A. (2014). The cloud. Salmagundi, 184, 129-143.
Cowls, J., Tsamados, A., Taddeo, M., & Floridi, L. (2021). The AI gambit: leveraging artificial intelligence to combat climate change—opportunities, challenges, and recommendations. AI and Society, 38(1), 283-307. https://doi.org/10.1007/s00146-021-01294-x
Dhar, P. (2020). The carbon impact of artificial intelligence. Nature Machine Intelligence, 2, 423-425. https://doi.org/10.1038/s42256-020-0219-9
Ensmenger, N. (2018). The environmental history of computing. Technology and Culture, 59(4), 7-33. https://doi.org/10.1353/tech.2018.0148
Montevecchi, F., Stickler, T., Hintemann, R., Hinterholzer, S. (2020). Energy-efficient Cloud Computing Technologies and Policies for an Eco-friendly Cloud Market. Retrieved June 6, 2023, from https://digital-strategy.ec.europa.eu/en/library/energy-efficient-cloud-computing-technologies-and-policies-eco-friendly-cloud-market
Negroponte, N., Harrington, R., McKay, S. R., & Christian, W. (1997). Being digital. Computers in Physics, 11(3), 261-262. https://doi.org/10.1063/1.4822554
Schwartz, R., Dodge, J., Smith, N. A., & Etzioni, O. (2020). Green AI. Communications of the ACM, 63(12), 54-63. https://doi.org/10.1145/3381831