Tackling energy and water challenges to navigate a sustainable path.
In a recent article by Kate Crawford, “Generative AI is guzzling water and energy”, she highlights that the environmental costs and impact of Generative AI have come under scrutiny. She sheds light on the excessive energy consumption and freshwater use of Generative AI, exemplified by models like ChatGPT. To address AI’s ecological impacts, she suggests that the “industry prioritise using less energy, building more efficient models and rethinking how it designs and uses data centres.” While access to accurate and complete data on the environmental impact of generative AI is hard to come by, legislators in the US have started taking notice. For example, on 1 February, the Artificial Intelligence Environmental Impact Act of 2024 was introduced. This Bill directs the National Institute for Standards and Technology to establish standards for assessing AI’s environmental impact in collaboration with academics, industry, and civil society and to create a voluntary reporting framework for AI developers, operators, and other measures.
In the South African context, where water is a scarce resource and an ongoing electricity crisis already plagues the country, this raises serious questions about the sustainability of using Generative AI and how the ecological impact of using AI can be limited or regulated. Therefore, the aim is to briefly explore the environmental challenges of Generative AI in South Africa and propose strategies and regulations to mitigate its ecological impact.
Energy consumption: A critical challenge
As also mentioned by Crawford, Generative AI demands a substantial amount of energy to execute searches and generate responses compared to conventional web searches. The energy-intensive nature of AI models raises alarms, especially in a country like South Africa grappling with an electricity crisis. The country already faces challenges in meeting its energy demands and grid constraints, which will intensify using generative AI.
Addressing this issue requires a multifaceted approach that combines technological advancements and responsible usage practices. One solution is to encourage the development of energy-efficient hardware for AI processing. For example, research and investment in low-power computing technologies can create more sustainable AI systems. Additionally, optimising algorithms to reduce computational complexity and exploring alternative energy sources for AI infrastructure can contribute to curbing the energy appetite of generative AI in South Africa.
Water scarcity: A compounding factor
In addition to energy concerns, the large quantities of freshwater required to cool AI’s processors pose another challenge in South Africa, where water scarcity is a pervasive issue. Given that water is already a scarce resource and that water shortages are widespread, it is imperative to rethink the water-intensive cooling methods employed in AI infrastructure. For example, adopting liquid cooling systems that utilise non-potable or recycled water may be considered an eco-friendly alternative, which may minimise the impact on the country’s freshwater resources.
Furthermore, promoting responsible and sustainable water usage practices in data centres and AI facilities can contribute to conserving this precious natural resource. To this end, implementing water recycling initiatives and investing in research for water-efficient cooling technologies are essential steps toward sustainable AI deployment in South Africa.
Policy and regulation: A framework for sustainability
Policy and law should be developed and implemented to address AI’s environmental impact and ensure the responsible and sustainable use of Generative AI in South Africa. Similar to the suggested Artificial Intelligence Environmental Impact Act of 2024, similar legislation should be considered in South Africa. For example, a ‘specific environmental management’ Act, which deals in more detail with specific aspects of the environmental impact of AI forming part of the overarching National Environmental Management Act 107 of 1998, and read in conjunction with the National Water Act 36 of 1998, could be formulated to provide for energy efficiency standards for hardware, firmware (algorithms) and data centres, water usage restrictions, compulsory public reporting (rather than voluntary) requirements and incentives for companies adopting eco-friendly AI practices.
For example, encouraging companies to adopt green AI practices through tax incentives and certifications, similar to Environmental, Social and Corporate Governance (ESG) reporting, can incentivise organisations and prioritise sustainability in their AI deployments. A regulatory, independent body, similar to the Information Regulator or ICASA, can also be established to support transparency and adherence to standards through conducting independent investigations and audits of AI developers and large-scale users, to support transparency. Adopting a transparent approach regarding the environmental impact of AI models and data centres can enhance accountability and encourage the industry to prioritise ecological sustainability.
Public-private collaboration
As part of a broader solution, another step would be to create awareness among AI developers and users about the ecological footprint of generative AI to foster a sense of collective responsibility.
Collective responsibility and collaborative efforts between the public and private sectors are required to significantly reduce the carbon footprint of generative AI in South Africa. Governments also play a pivotal role by providing financial support for research and development initiatives focused on sustainable AI technologies. Collaboration between tech companies, research institutions, and environmental organisations can lead to the creation of innovative solutions that balance the benefits of AI with environmental preservation. Furthermore, industry leaders should proactively invest in green technologies and commit to sustainable practices in their AI operations.
Conclusion
As highlighted by Crawford’s article, the environmental challenges associated with generative AI necessitate a strategic and collaborative approach to ensure sustainability in South Africa. Addressing AI models’ energy consumption and water-use requirements is paramount, especially in a country facing an electricity crisis and water scarcity. South Africa can navigate the delicate balance between technological advancement and environmental preservation by developing energy-efficient hardware and firmware, responsible water usage practices, robust laws and policies, and public-private collaboration. By embracing eco-friendly AI practices, the nation can harness the potential of generative AI without compromising its commitment to a sustainable and resilient future.
Dr Tina Gama-Kotzé is the Research and Didactic Lead in Law & Ethics at Boston City Campus. She is a Y2-rated researcher and has experience in face-to-face and online higher education. She holds a BA (Law), LLB, LLM (cum laude) and LLD from Stellenbosch University. Her own scholarship focusses on the interplay between property rights, environmental law and energy law in the constitutional dispensation.