Governments worldwide have been adopting national AI strategies to guide the design of policies for enhancing Artificial Intelligence (AI) development. Through these strategies, many countries have expressed AI’s potential for helping combat environmental degradation, but almost none have reflected on the role of strategies in reducing AI’s own environmental impacts. This article discusses this trend and reflects on how public policies could help make AI development and deployment less impactful.
The advancement of the digital industry in the last few decades has held a promise that digitising the world would be an essential part of the pack of responses for reducing environmental degradation. Artificial intelligence technologies, in particular, have been advocated by many as a key component in promoting sustainable development.
However, as the proliferation of these technologies has intensified, their environmental impacts have also grown. While digital systems, and particularly AI, can indeed support combating climate change, the development, usage and discarding of digital systems and hardware have been increasing water, energy, mineral and gas consumption rates, as well as e-waste dumping.
Addressing AI’s environmental footprint requires a multifaceted approach. Regulation is one of them. It could, for instance, help promote further transparency from the industry, especially considering that companies do not, so far, have specific obligations to disclose information on the consumption of the resources to develop their AI systems – although some countries request information, for instance, on mineral usage from zones in conflict or energy expenditure in data centres, but not specifically on the digital systems’ impacts per se. It could also establish standards for resource consumption, and strengthen rules that already exist for inhibiting the trade of minerals from zones of conflict or Indigenous territories, along with other measures.
However, an effective response towards a more sustainable AI development also requires, beyond reactive measures, positive action. If implemented hand in hand with regulatory efforts, national AI strategies, as policy instruments for promoting AI development within countries, can play a strategic role.
The role of national AI strategies
National AI strategies have been issued in the last years by countries around the world to guide the design of policies for AI development. Usually, these documents establish recommendations, milestones and metrics related to capacity building, ethical standards and funding for AI development, among others.
If we analyse these strategies through the lens of environmental sustainability, the German strategy, for instance, incentivises the development of systems that support reducing greenhouse gas (GHG) emissions and highlights the need for an assessment of the necessity to improve environmental regulations for promoting sustainable AI systems. The strategy focuses on energy reduction, which may reflect the German government’s worry regarding the carbon footprint of its energy system, which is still largely dependent on fossil fuels.
The Chinese strategy, in turn, calls for actions that promote ‘intelligent environmental protection’, such as by constructing intelligent prevention and control systems for environmental protection and sudden environmental events.
These calls relate to what has been called AI for sustainability, namely the application of AI systems to achieve sustainability, which does not necessarily relate to the sustainability of AI (promoting the sustainability of AI systems per se).
However, as Perucica and Andjelkovic highlight in an analysis of national AI strategies published by European countries, Japan and China prior to 1 June 2021, ‘with the exception of France and Japan, almost no strategy recognises that the development of sustainable AI solutions are only feasible if those solutions are sustainable by design’. In other words, national strategies seem to be more worried about incentivising AI for sustainability instead of the sustainability of AI.
The same issue can be identified in countries in the Majority World. In Latin America, two examples are the AI strategies of Brazil (now under review) and Chile. In both cases, the high resource consumption is not mentioned in the mapping of risks posed by AI. That despite the fact that Brazil has been the source of resources such as gold and cassiterite, particularly from Indigenous territories, to feed the global digital industry, and Chile the source of lithium and water, although recent social movements have provided resistance to data centre projects in the country.
This scenario is particularly problematic if we consider that national AI strategies represent how the public sector aims to promote AI development within their territories and constitute a hybrid of policy and discourse that shapes how narratives guide the development of public policies in this area. But what exactly are these strategies thus missing?
An environmental chapter in AI strategies?
The complex distribution of the AI industry’s environmental impacts in terms of materials and territories makes it a challenging issue to be tackled. Below I mention some possible actions to be taken, within a list that not only has to be further detailed and extended, but also considered within the limitations that the digital policy field in isolation entails. These limitations relate both to an inevitable need to concert efforts with other policy areas – such as energy, for example – and the fact that the digital industry lies within a capitalist dynamic, which is historically colonial and, hence, extractivist. It is thus part of a broader, structural problem, which the ideas mentioned below might merely mitigate, but not solve.
One of the actions concerns infrastructure. When planning the building of data centres, for instance, governments should invest in solutions that optimise resource consumption and focus construction of these centres in areas where water is more abundant and energy sources are the least impactful as possible. It is worth noting, however, that focusing on renewable energy may reduce impact, but is not a panacea. To mention an example, the construction of hydroelectric power stations can damage whole ecosystems, impacting fauna, flora and local economies. The same applies to the building of wind farms, which has had impacts in Quilombola communities’ access to water in Brazil.
Policies for increasing the lifetime of hardware are also relevant, both for decreasing the demand for minerals and for reducing e-waste. Brazil and the European Union have been among the jurisdictions discussing this issue at a legislative level as a so-called right to repair. In India, the executive branch has been leading initiatives for connecting consumers with producers for repairing devices. Strategies could follow this initiative, contributing further to this domain through policies for both repairing and recycling these materials. It is key, however, that these policies not only reduce disposal but also make governments responsible for their own detritus, including by inhibiting exports of debris to other countries (a practice that Europe has had with Ghana, for instance).
Strategies should prioritise investments towards developing AI systems that help tackle environmental degradation, but only insofar as these systems are also sustainable per se, and fulfil at least the sufficient conditions to contribute to environmental protection. Beyond AI solutions that help predict disasters or contribute to optimising resource usage, attention should also be given to community uses of AI for protecting their territories. For example, some Indigenous leaders in Brazil have been asking for funds for purchasing drones to patrol their lands against illegal gold miners and other agents of deforestation. AI strategies could provide methodologies for assessing these demands and correlate them with the work of environmental agencies, and methods for allowing state or Indigenous sovereignty over the systems and data.
Building capacity within communities with regard to developing, deploying and thinking about these systems, as well as learning from these groups about how to appropriate them, is also crucial to improving the understanding of both communities and policymakers on the impact they have. This is a key step for making those systems that we, as a society, most desire, flourish, without falling into technosolutionist traps. At the same time, meaningfully engaging and listening to local communities, including Indigenous peoples, for their essential role in environmental protection is fundamental.
In addition, investment should be driven towards enhancing quantification and reduction of resource consumption. This could even involve universities, companies, standardisation bodies and public agencies in a concerted effort to create techniques and parameters that enhance the sustainability of AI systems.
Finally, it is key that the whole AI and digital ecosystem enhances its transparency, especially on resource consumption, a theme to also be dealt with through regulation.
The need for concerted policy efforts
National strategies, for their role in centralising methods for policy design in the field of AI, could be effective drivers of further environmental sustainability in the industry. However, they cannot be planned in isolation. It is key to involve policy areas such as energy planning, environmental and animal protection, education, and law enforcement. National AI strategies could help concert these efforts in the digital policy sphere, and connect to these other sectors. But the challenge is significant: AI policies can only be effective to the extent that all these areas are jointly engaged in a serious environmental compromise.
This reflects the fact that AI’s materiality involves industries that are deeply rooted in dynamics of exploitation arising from capitalism and neo-colonial practices, including mining, energy, water and waste dumping. A comprehensive, multifaceted approach that goes beyond AI towards the digital industry and the overall economic dynamics needs thus to be considered, so as to improve transparency and promote a more equitable distribution of AI’s benefits.
Moreover, this is even more important when one considers that the digital industry’s demand is expected to rapidly increase. Although the digital industry has been driving significant efforts to reduce resource demand, particularly concerning data centres, it is still unclear if these efforts will advance fast enough with regard to the increasing demand of AI systems. Preventive action needs thus to be adopted immediately so we do not rely on an unsustainable industry to build a new layer of infrastructure in our societies, in a way that focuses on protecting not the interests of a few, but the livelihood of other species, landscapes and communities.
