AI and the Sustainable Development Goals

Unveiling Hidden Costs and Exploring Potential

Taufik Muhamad Yusup

Artificial Intelligence is often believed to be the modern-day steam engine, fueling a new era of innovation and economic growth. This claim makes sense, as industries race to develop AI technology, frequently mentioning ‘AI’ whenever showcasing new breakthroughs to stay relevant in this hypedriven environment. AI has also become an integral tool in everyday life. For instance, in academia, we often use AI chatbots to help us understand complex concepts more easily or to make the research process more efficient. This increase in efficiency and productivity is not limited to academia; it spans various sectors, driving significant economic impact. According to PricewaterhouseCoopers (PwC), a global professional services network known for providing audit, assurance, consulting, and tax services, AI has the potential to significantly transform productivity and GDP growth across the global economy, with a projected contribution of $15.7 trillion by 2030¹. However, behind its remarkable capabilities and widespread applications lie significant hidden costs that we often do not realize. These costs are not just financial but also environmental and ethical, raising questions about the true sustainability of this technology.

As we increasingly rely on AI to drive progress, it becomes crucial to examine its alignment with our global sustainability goals, particularly those outlined in the United Nations’ Sustainable Development Goals (SDGs). The SDGs set an ambitious agenda for achieving a better and more sustainable future for all by 2030, and AI has the potential to play a significant role in reaching these targets. However, while AI can contribute significantly to these goals, it also poses challenges that may conflict with them, particularly in terms of environmental and ethical costs. To fully assess AI’s impact on the SDGs, we must consider both its benefits and the hidden costs associated with its development and deployment. In this article, I aim to uncover the unseen aspects of AI, exploring both the risks and opportunities it presents in addressing critical sustainability challenges.

Tangibility and the Hidden Cost of AI

Most of us would probably think that AI is intangible, as we usually interact with it through an interface that can be accessed using our computer—it is something we cannot touch unless it is embodied in a physical form like a robot. Furthermore, when people discuss AI, they most likely talk about models such as deep neural networks, large language models (LLMs), and generative adversarial networks (GANs). However, people are usually not aware of what AI was made from, which actually relies heavily on physical components. Unfortunately, extracting these tangible building blocks of AI requires sacrifices from Mother Earth, often in ways that are not sustainable.

Data and The Energy It Takes

Data is the foundation of artificial intelligence without vast amounts of data, AI would be unable to function. In fact, data is the key to enabling AI systems to learn, make decisions, and evolve. Electricity Consumption Imagine data as the lifeblood of AI, with data centers acting as its heart, pumping vast amounts of information through complex systems to bring AI to life. However, this vital relationship comes with significant energy demands that are often overlooked.

• Electricity Consumption

For example, every time we do something as simple as a Google search, it consumes about 0.3 watt-hours of electricity—comparable to keeping an LED light bulb on for 10 seconds. More advanced queries to AI tools like ChatGPT require approximately 2.9 watt-hours per request, equivalent to leaving the same light bulb on for almost 2 minutes. With 9 billion Google searches conducted daily, switching all those to AI-powered models like ChatGPT would increase electricity demand by almost 10 terawatt-hours (TWh) annually, enough to power over a million U.S. households for a year.²

When we look at Earth from space at night, as captured in satellite imagery, the bright lights illuminating Japan are a striking sight. This constant glow represents a massive energy consumption. Now, consider that the energy required to keep those lights shining is equivalent to the electricity consumption projected for data centers worldwide by 2026. According to a recent International Energy Agency (IEA) report, data centers’ total electricity consumption could reach more than 1,000 terawatt-hours (TWh) by then, roughly matching Japan’s entire electricity use⁴. As AI services continue to grow, driven by power-hungry GPUs, the energy demands of data centers will only increase, underscoring the need for sustainable solutions.

• Carbon Emissions

When people say data is the new oil, I believe it holds true not only in terms of value but also in its contribution to carbon emissions. Data centers and data transmission networks are responsible for about 1% of global energy-related greenhouse gas (GHG) emissions6. This is because the energy consumed by IT equipment—such as servers, routers, and storage devices—directly correlates to a data center’s carbon footprint, which refers to the total amount of greenhouse Carbon Emissions Water Usage gas emissions produced by its operations. This footprint includes emissions from energy consumption, cooling systems, and other infrastructure required to keep data centers running. As the data load increases with our use of digital services—whether for streaming, searching, or storing data—so does the energy demand, which drives up these facilities’ carbon footprint. The process of how data centers can contribute to GHG emissions can be seen in Figure 2. In 2020, data centers and networks accounted for approximately 330 million metric tons of CO2 equivalent which roughly the same amount of carbon emitted by 72 million cars driven for an entire year.⁶

Another concerning aspect of data’s environmental impact is the phenomenon known as ‘dark data.’ This term describes the vast amounts of digital information that are stored in the cloud but rarely, if ever, accessed again. To put this into perspective, consider a typical company in sectors like insurance, retail, or banking with 100 employees. Such a business might generate around 3,000 gigabytes of dark data daily. If this data were kept for a year, its carbon footprint would be comparable to taking six round-trip flights from London to New York.⁷

• Water Usage

Tech companies need to ensure that data centers remain cool enough to operate 24/7 without overheating. This is typically achieved through cooling systems that rely heavily on water to regulate temperature. These systems work by using large volumes of water to dissipate the heat generated by the constant operation of data centers, which manage the storage, processing, and transmission of vast amounts of data. Unfortunately, companies are usually not transparent about the resources needed to achieve this.

While the carbon footprint of artificial intelligence (AI) models, particularly large ones like GPT-3, has received significant attention, the substantial water footprint of these models has often been neglected. For instance, training GPT-3 at Microsoft’s advanced U.S. data centers can result in the evaporation of 700,000 liters of clean freshwater, a fact that is not widely publicized.⁹

Looking ahead, the global demand for AI could require between 4.2 to 6.6 billion cubic meters of water annually by 2027. This amount surpasses the total yearly water usage of Denmark and is nearly half of that used by the United Kingdom.¹⁰ This issue is especially pressing given the increasing freshwater scarcity driven by a growing global population, diminishing water supplies, and outdated water infrastructure.

Extracting the Earth to Build the Skeleton of AI

I did not fully realize the extent of the environmental impact involved in building AI until I read “Atlas of AI” by Kate Crawford. The tangible components of AI, such as servers, GPUs, and other hardware, rely heavily on extracting minerals and metals from the earth. Regrettably, this extraction process is often ignored in discussions about AI. Mining for these essential materials involves significant environmental and human costs.

For instance, in her book, Crawford mentioned my country, Indonesia, as the second-largest metal producer after China. She specifically highlighted the national tin corporation, PT Timah, since tin is one of the elements needed to create semiconductors. The tin mining by PT Timah, conducted on the islands of Bangka and Belitung, has dramatically altered the island’s landscapes, ruined its farms and forests, killed its marine life, and negatively impacted tourism.¹² This fact resonates with me, as Indonesia’s mining industry has long been a topic of national discussion. While it’s a source of economic strength, it also comes with a heavy environmental cost. Hearing about the impact of tin mining on islands like Bangka and Belitung makes me reflect on how pursuing technological advancement often comes at the expense of our natural ecosystems. Additionally, in the first semester of 2024, there was a significant corruption case related to PT Timah, causing state losses of 271 trillion rupiah,¹³ which became a hot topic among the public.

Furthermore, the environmental impact extends beyond tin mining. GPUs, for instance, are composed of silicon layered with tantalum and palladium transistors and capacitors for better storage on a smaller chip. They incorporate a mind-boggling array of chemicals and elements like copper, boron, cobalt, and tungsten. These rare earth metals, primarily sourced from Southeast Asia, are known as conflict materials due to their involvement with unjust labor practices and their high environmental impact, including the release of greenhouse gases thousands of times more potent than CO2.¹⁵ This is largely due to a lack of regulation in the rare earth metal mining industry. Additionally, the production process of these metals is highly toxic. Unfortunately, countries like China, and others with low environmental standards, have thriving illegal gem and metal mining sectors linked to rainforest depletion and the destruction of natural habitats. This problem is not isolated to a single country but is an industrywide issue. The environmental impact of mining rare earth metals to manufacture graphics cards is extremely high, with greenhouse gases making up around 50% of emissions. There is an urgent need to manage these impacts effectively.

AI as a Catalyst for Sustainability

While the development and deployment of AI come with significant environmental and ethical costs, AI also holds immense potential to contribute positively to sustainability efforts. By leveraging AI technology, we can address some of the most pressing environmental challenges, optimize resource use, and support the United Nations’ Sustainable Development Goals (SDGs). In fact, The United Nations has actively explored the role of AI through its 2023 United Nations Activities on Artificial Intelligence (AI) Report, which was developed in collaboration with 46 UN agencies. This report highlights 408 AI projects that contribute to all 17 SDGs, ranging from forecasting food crises and monitoring water productivity to mapping schools via satellite imagery and optimizing communication networks.¹⁶ These initiatives showcase how AI is being used globally to promote sustainability and address critical challenges across sectors.

We might be familiar with AI applications in business, enhancing productivity and driving economic growth. However, AI’s potential extends far beyond business. Currently, I am taking a summer course online from Climate Change AI,¹⁷ which has opened my eyes to the vast and varied ways AI can contribute to sustainability. For example, I was particularly struck by how AI is being used to improve energy efficiency in smart grids, predict extreme weather events with reasonable accuracy, and even monitor biodiversity by analyzing large datasets from remote sensing technologies. These real-world applications demonstrated to me the transformative power of AI in addressing environmental challenges, reinforcing the importance of integrating AI into our sustainability efforts. I have thoroughly enjoyed the course, as it not only provided me with new knowledge but also included technical tutorials on how AI can be practically applied in these areas. Additionally, the book AI in the Wild: Sustainability in the Age of Artificial Intelligence by Peter Dauvergne provides compelling examples of AI’s role in promoting sustainability.

AI and Climate Change

AI can be instrumental in climate change mitigation efforts, supporting SDG 13 (Climate Action). One interesting example of AI in climate action is its ability to forecast rare extreme weather events accurately. AI models have shown proficiency in predicting tropical storms and hurricanes making landfall on the western seaboard of the United States, which are considered extremely rare. For instance, the AI model forecast for a tropical storm on August 18, 2023, closely matched the actual conditions, outperforming traditional forecasting methods.¹⁸

Furthermore, AI models such as MetNet2 are highly effective in nowcasting severe convective events up to 12 hours in advance. These models combine convolutional neural networks (CNNs) and long short-term memory (LSTM) modules to provide accurate short-term forecasts, which are essential for preparing for and mitigating the impacts of severe weather. These AI weather models, including state-of-the-art approaches like FourCastNet by NVIDIA and GraphCast by Google Research/DeepMind, have shown great promise in providing faster and more accurate weather predictions than conventional numerical weather prediction (NWP) models.

In addition to these advancements, I am conducting research incorporating AI for climate. I am working on predicting paleo precipitation anomalies using deep learning techniques. This research aims to improve our understanding of historical climate patterns and enhance our ability to predict future climate changes, providing valuable insights for climate mitigation and adaptation strategies. By leveraging deep learning, we hope to uncover patterns and trends in past precipitation data that can inform more accurate and reliable climate models, ultimately aiding in the development of effective climate policies and interventions.

Protecting Biodiversity

AI can also contribute significantly to biodiversity conservation, supporting SDG 15 (Life on Land). By analyzing data from satellite images, drones, and camera traps, AI can help monitor and protect endangered species and their habitats. One fascinating example from the Climate Change AI course is the use of camera traps combined with AI to monitor wildlife. Camera traps are stationary sensors that capture high-resolution images and videos of wildlife in their natural habitats. These cameras can collect vast amounts of data, which AI models such as the MegaDetector then analyze to identify and track individual animals, detect poaching activities, and monitor population changes. For example, AI-powered camera traps have been used in the Serengeti to track the movements and behaviors of various species, providing invaluable data for conservationists.²⁰ This approach not only improves the accuracy and efficiency of wildlife monitoring but also allows for real-time responses to potential threats.

Moreover, AI can assist in combating overfishing, supporting SDG 14 (Life Below Water). To understand why and where overfishing is occurring, a team of researchers is using machine learning to assess the “global footprint” of fishing. David A. Kroodsma, director of research at Global Fishing Watch, led a study that found industrial fleets were fishing across more than half of the world’s ocean area. Using convolutional neural networks (a deep learning method) to identify vessel features and fishing activity, the team tracked more than seventy thousand fishing vessels from 2012 to 2016 by analyzing 22 billion satellite and land-based positioning messages from the automatic identification system (AIS).²¹ This comprehensive tracking helps detect illegal fishing practices and ensure sustainable fishing quotas are met, preserving marine ecosystems and ensuring the long-term sustainability of fish populations.

Optimizing Energy Consumption

AI also plays a crucial role in optimizing energy consumption, aligning with SDG 7 (Affordable and Clean Energy). AI algorithms can enhance the efficiency of power grids by predicting energy demand and optimizing electricity distribution, reducing waste, and ensuring that renewable energy sources are utilized effectively.

One prominent example from the Climate Change AI course involves AI-driven energy forecasting models. These models can predict day-ahead solar irradiance using Vision Transformers (ViTs)16, which are critical for day-ahead commitments to the grid and operational planning. By accurately forecasting solar power generation, AI helps grid operators balance supply and demand more effectively, reducing the reliance on fossil fuels and increasing the integration of renewable energy sources.

Can AI Help Us Achieve the UN SDGs by 2030?

AI has both positive and negative impacts on sustainability. To answer the question of whether it can help us achieve the SDGs by 2030, it ultimately depends on us.

Achieving the UN SDGs by 2030 is a highly ambitious goal, given that we have less than six years to make significant progress. While AI offers promising solutions, immediate and concerted efforts are required to harness its potential effectively. Policymakers, industry leaders, and researchers must collaborate to address several challenges.

Firstly, the environmental costs associated with AI development and deployment cannot be overlooked. High energy consumption, significant carbon emissions, and extensive water usage are critical issues that need to be managed. To address these challenges, we must invest in renewable energy sources to power data centers and AI infrastructures. This could involve adopting more energy-efficient hardware, utilizing energy-saving algorithms, and deploying AI models in regions where renewable energy is readily available. Additionally, AI-driven tools could help monitor and optimize energy use in real-time, reducing unnecessary consumption.

Secondly, the extraction of minerals and metals required for AI hardware has substantial environmental and human impacts. To mitigate this, governments and companies could collaborate to enforce stricter regulations on ethical mining practices, ensuring materials are sourced with minimal environmental disruption and fair labor conditions. Encouraging the recycling and reuse of electronic components could significantly reduce the demand for newly mined materials. Companies could also adopt thirdparty certification programs, similar to “Fair Trade” in agriculture, where independent organizations audit and certify that materials are ethically sourced and environmentally sustainable.

Moreoever, transparency and accountability in the tech industry are crucial. Governments could create regulations that require companies to publicly disclose their environmental impact, with penalties for non-compliance. Another idea is to develop sustainability scorecards for companies, which would be publicly accessible and regularly updated by independent environmental agencies. These scorecards could provide consumers and investors with clear insights into a company’s energy usage, carbon emissions, and water consumption, encouraging better environmental practices. Additionally, companies could be incentivized to adopt “green AI” solutions by receiving subsidies or tax benefits for lowering their carbon footprints.

To fully realize the potential of AI in achieving the SDGs, a concerted effort is needed to implement policies that promote sustainability and accountability. This involves developing and enforcing regulations to ensure the sustainability of AI technologies, promoting transparency in the tech industry, investing in eco-friendly AI solutions, and fostering international cooperation to share knowledge and best practices.

In conclusion, whether AI can help us achieve the UN SDGs by 2030 depends on how quickly and effectively we can address its environmental and ethical costs. The potential of AI to drive progress towards these goals is clear, but realizing this potential requires immediate and collaborative efforts from all stakeholders. With the right policies and cooperation, we can ensure that AI becomes a force for good, driving us towards a more sustainable and equitable future. The urgency to act is clear, and the time to implement sustainable AI practices is now.

---

References

1「Sizing the prize PwC’s Global Artificial Intelligence Study: Exploiting the AI Revolution What’s the
real value of AI for your business and how can you capitalise? 」 ,
https://www.pwc.com/gx/en/issues/data-and-analytics/publications/artificial-intelligence-study.html

2 「 How Much Data Do We Create Every Day? The Mind-Blowing Stats Everyone Should Read 」 ,
https://www.forbes.com/sites/bernardmarr/2018/05/21/how-much-data-do-we-create-every-day-themind-blowing-stats-everyone-should-read/

3「Japan at Night」, https://earthobservatory.nasa.gov/images/147987/japan-at-night

4「 IEA Study Sees AI, Cryptocurrency Doubling Data Center Energy Consumption by 2026 」 ,
https://www.datacenterfrontier.com/energy/article/33038469/iea-study-sees-ai-cryptocurrencydoubling-data-center-energy-consumption-by-2026

5「How to make digital infrastructure sustainable by tackling greenhouse gases: Uptime Report」,
https://www.dailyhostnews.com/how-to-make-digital-infrastructure-sustainable-by-tacklinggreenhouse-gases

6「 Data Centres and Data Transmission Networks 」 , https://www.iea.org/energysystem/buildings/data-centres-and-data-transmission-networks

7「 What is ‘dark data’ and how is it adding to all of our carbon footprints? 」 ,
https://www.weforum.org/agenda/2022/10/dark-data-is-killing-the-planet-we-need-digitaldecarbonisation/?
utm_campaign=social_video_2022&utm_content=27617_dark_data&utm_medium=social_video&ut
m_source=linkedin&utm_term=1_1

8「 Cooling Water Efficiency Opportunities for Federal Data Centers 」 ,
https://www.energy.gov/femp/cooling-water-efficiency-opportunities-federal-data-centers

9 Crawford, K. (2024), “Generative AI’s environmental costs are soaring — and mostly secret”, Nature,
Vol. 626(8000), p. 693. https://doi.org/10.1038/d41586-024-00478-x

10 Li, P., Yang, J., Islam, M. A., & Ren, S. (2023), “Making AI Less ‘Thirsty’: Uncovering and Addressing the
Secret Water Footprint of AI Models”, arXiv.org. https://arxiv.org/abs/2304.03271

11「 The Contribution of Bangka Belitung Tin is Imbalanced by Its Corruption Value 」 ,
https://www.kompas.id/baca/english/2024/04/03/en-kontribusi-timah-bangka-belitung-timpangdengan-nilai-korupsinya

12 Crawford, K. (2021), Atlas of AI, New Haven: Yale University Press.

13「 PT Timah Corruption: State Losses IDR 271 Trillion Due to Environmental Damage 」 ,
https://asiatoday.id/read/pt-timah-corruption-state-losses-idr-271-trillion-due-to-environmentaldamage

14「Nvidia GPU — Design Life-Cycle」, https://www.designlife-cycle.com/nvidia-gpu

15「 Environmental Implications of the AI Boom 」 , https://towardsdatascience.com/environmentalimplications-of-the-ai-boom-279300a24184

16「UN AI Actions – AI for Good」, https://aiforgood.itu.int/about-ai-for-good/un-ai-actions/

17 https://www.climatechange.ai/

18 Mitra, Peetak. “An Introduction to AI for Weather Forecasting.” Climate Change AI Summer School
2024: AI for Weather, 2 July 2024, Online. Lecture.

19 Espeholt, L., Agrawal, S., Sønderby, C., Kumar, M., Heek, J., Bromberg, C., Gazen, C., Carver, R.,
Andrychowicz, M., Hickey, J., Bell, A., & Kalchbrenner, N. (2022), “Deep learning for twelve hour
precipitation forecasts”, Nature Communications, Vol. 13(1). https://doi.org/10.1038/s41467-022-
32483-x

20 Tuia, Devis. “Machine Learning for Animal Conservation: From Sensors to Knowledge.” Climate
Change AI Summer School 2024, 27 June 2024, Online. Lecture.

21 Dauvergne, P. (2020), AI in the Wild: Sustainability in the Age of Artificial Intelligence, Cambridge, MA:
MIT Press