The Environmental Costs of Generative Artificial Intelligence
A guest post by Marco Antonio Lueras, New Mexico Conservation Alliance.
Hello Everyone,
With so much focus on the energy demands of the future of datacenters and future quantum hybrid Fabs of the world, when you think of about all those A.I. chips and those massive AI supercomputers that will come into being, what about the water?
Recently on the whole, I’m fairly obsessed with whether AI will accelerate or delay the race to net-zero emissions? As artificial intelligence transforms the global economy, there might be a fairly significant ecological cost. How responsible in ESG terms will the AI hyperscalers truly be?
NPR reports the average data center uses 300,000 gallons of water a day to keep cool, roughly equivalent to water use in 100,000 homes. I wanted to find somehow who might know a thing or two about this on a deep academic and sociological basis. Luckily I found Marco.
So what happens when datacenter greed begins to intersect with droughts in the real world? With drought spreading around the globe, battles over water are erupting between AI companies seeking more computing power and communities where their facilities are located.
This is only the beginning. In this post Generative artificial intelligence will appear as the acronym GAI. The Cloud, datacenters and AI appear to be on a collision course with climate change more frequent droughts and water scarcity.
The world will need radical solutions to the energy, water, environmental and the economic demands for AI datacenters, supercomputers and the like that appears to be on the horizon.
AI’s Carbon Footprint is going to be Problem
Over the past decade, the compute capacity used to train advanced large language models has increased tenfold each year. Demand for AI services is expected to rise by 30–40% annually over the next 5–10 years. And more powerful AI models will require more energy. One estimate suggests that, by 2027, global AI-related energy consumption could be 10 times greater than it was in 2023.
Support and Dive Deeper
Support my work for as little as $2 a week. 🎓📚💡
Marco’s Articles to Read
Marco specializes in Responsible AI at the intersection of water conservation. An avid research pioneer, Marco serves on several advisory boards for companies working in the intersections of AI and water.
By Macro Antonio Lueras, written in March, 2024.
The Environmental Costs of Generative Artificial Intelligence
Article Summary:
The development of artificial intelligence (AI) is resource and energy intensive. At the same time, it has been touted as an effective tool against climate change. This article investigates the environmental costs of GAI and presents a set of sustainable recommendations for the AI community to consider.
If you believe the environmental impact of AI is or might become a big deal, please consider sharing this article.
The Environmental Costs of Generative Artificial Intelligence
Created in IZEA (AI generated)
AI Optimism
The world as it used to be is fading into obsolescence. The AI boom over the past decade has produced some of the most readily accessible and innovative pieces of technology humankind has ever seen. It is safe to say that generative AI has solidified itself into the mainstream economy. The CEO of Tesla, Elon Musk, has even gone so far as to say:
“We will have for the first time something smarter than the smartest human. It's hard to say exactly what that moment is, but there will come a point where no job is needed.” (Forbes)
CEOs, founders, and other entrepreneurs have been raving about generative AI for the past few years. Despite Musk’s and others’ optimism, the promise of technology that transcends human intelligence is a promise that has gone unanswered. Musk’s remarks harken back to the 1960s, when AI pioneer Herbert Simon discussed the human-technology relationship:
“Technologically, as I have argued earlier, machines will be capable, within twenty years, of doing any work that a man can do. Economically, men will retain their greatest comparative advantage in jobs that require flexible manipulation of those parts of the environment that are relatively rough — some forms of manual work, control of some kinds of machinery (e.g., operating earth-moving equipment), some kinds of nonprogrammed problem solving, and some kinds of service activities where face-to-face human interaction is of the essence.” (Open Philanthropy)
When speaking of interaction and a computer’s role in decision-making, it is difficult (perhaps even impossible) to draw a line between what counts as a problem of purely logical reason and what does not. Even with the rise of the innovative “expert knowledge systems” of the 1980’s, Simon’s predictions went unfound. In an effort to describe what human knowledge is, MIT philosopher Hubert Dreyfus and his brother, Stuart Dreyfus, a RAND corporation scientist, wrote a landmark book in the 1970’s that criticized the limits of artificial intelligence. The Dreyfus brothers’ work influenced the likes of Terry Winograd and Joseph Weizenbaum, two pioneers in the field of computer science. Skepticism of AI accordingly grew in the latter half of the twentieth century.
Despite such criticism and unfound predictions, computer science has continued to revolutionize humankind’s understanding of what it means to interact, to belong, and to be human. Long gone are the days when Joseph Weizenbaum solemnly lamented about the future of AI. So too are the days when skeptics commanded the fate of AI. In 2024, business optimism for the future of generative AI remains at an all time high. This optimism has fueled a boom in data center production and chip fabrication contracts. But, the American public is not so sure:
Figure 1. Pew Research Center Study
In a three year survey of American’s views on artificial intelligence, the Pew Research Center found that concern for the use of AI in daily life has grown by 15%. At the same time, excitement for the use of AI in daily life has dropped by 8%. If CEOs are “all in” on AI, why does the American public seem so unsure? There are three important hypothesized reasons:
(1) Economic displacement
(2) Climate change exacerbation
(3) Human rights
The second reason is the focus of the current conversation. Important to consider is the fact that world governments are slowly slipping away from reaching their climate goals heading into the latter half of the 2020s.
The Global Water Crisis
Water is life, but safe water is limited. According to a recent UNESCO press release, 2 billion people (26% of the world’s population) do not have safe drinking water and 3.6 billion (46%) lack access to safely managed sanitation. Half of the world’s population could be living in areas facing water scarcity by as early as 2025 (UNICEF). By 2040, roughly 1 in 4 children worldwide will be living in areas of extremely high water stress. The time for action is now.
The global water crisis can be defined as a multifaceted issue characterized by water scarcity, pollution, inadequate access to clean water, and unequal distribution of water resources. Key factors contributing to this crisis include population growth, urbanization, industrialization, climate change, and inefficient water management practices.
Many regions around the world face severe water scarcity, leading to increased competition for limited water resources among various sectors such as agriculture, industry, and households. This scarcity is exacerbated by pollution from industrial waste, agricultural runoff, and untreated sewage, rendering many water resources unusable or hazardous to human health.
Access to clean water is a significant challenge for billions of people, particularly in developing countries, where infrastructure for water supply and sanitation is inadequate. Lack of access to clean water and sanitation facilities leads to numerous health problems, including waterborne diseases such as cholera, diarrhea, and dysentery, contributing to high morbidity and mortality rates, particularly among children.
The unequal distribution of water resources exacerbates social and economic disparities, leading to conflicts over water rights and access, both within and between countries. The fight for the Colorado River in the US best exemplifies this. Climate change further intensifies the water crisis by altering precipitation patterns, increasing the frequency and severity of droughts and floods, and affecting the availability of water resources.
Addressing the global water crisis will require coordinated efforts at the local, national, and international levels. Sustainable water management practices, investment in infrastructure for rechargeable water supply and sanitation, promotion of water conservation and efficiency measures, and regulatory policies to mitigate pollution are essential components of any comprehensive solution.
Promoting equitable access to clean water and addressing the root causes of water scarcity and inequality are crucial for achieving long-term sustainability and ensuring the well-being of both people and the planet.
Establishing conservation-focused initiatives in the U.S. is important to successfully secure water for generations to come. For a “quick facts” video review about the global water crisis, watch this video from Netflix. As the world sinks further into the global water crisis, new technologies must emerge to help humankind cope with an uncertain future.
Environmental Costs of GAI
Will AI be the savior in a doomsday water scenario? Some claim yes, while others claim no. It is important to list and describe the potential environmental costs of generative AI:
Increased energy consumption
Increased carbon emissions
Data center production and infrastructure
Resource depletion and e-waste
Indirect impacts on other sectors
Figure 2. The Environmental Costs of GAI
Let’s discuss energy consumption and carbon emissions first. Training large language models (LLMs) like ChatGPT involves iterative processes of feeding massive amounts of data through complex neural networks, which requires extensive computational power. This computational workload demands energy-intensive hardware, such as Graphics Processing Units (GPUs) or Tensor Processing Units (TPUs), operating at high capacities for extended periods of time. As a result, data centers hosting these models consume substantial amounts of electricity, often drawing from non-renewable sources, thus contributing to carbon emissions and environmental strain.
The energy-intensive operations of data centers, compounded by the use of fossil fuel-based electricity, contribute significantly to carbon emissions. The carbon footprint of training and running generative AI models is substantial, with estimates suggesting that training a single large model can emit as much carbon dioxide as several cars over their lifetimes. This reliance on non-renewable energy sources exacerbates climate change, impacting ecosystems and biodiversity worldwide. Regarding water footprint, a recent study found that ChatGPT utilizes 500 milliliters of water every time you ask it a series of 5 to 50 prompts or questions.
Increased data center production and infrastructure also poses an important environmental cost. The physical infrastructure supporting data centers, including server racks, cooling systems, and networking equipment, combine to make up significant environmental costs. Constructing and maintaining these facilities require vast amounts of resources, including land, concrete, steel, and water. Additionally, the energy-intensive cooling systems necessary to dissipate the heat generated by servers contribute to further energy consumption and environmental strain.
Concerning resource depletion and e-waste, is it important to consider the fact that technology is a global economic machine. The production of hardware components for AI infrastructure relies on the extraction of natural resources, including metals, minerals, and rare earth elements. Mining and processing these resources can have significant environmental impacts, such as habitat destruction, soil and water pollution, and biodiversity loss. The finite availability of certain critical materials may lead to resource scarcity and geopolitical tensions. As generative AI technology evolves, older hardware used for training and deploying models may become obsolete, leading to electronic waste (e-waste). Disposal of outdated servers, GPUs, and other computing equipment poses environmental challenges due to the presence of hazardous materials such as lead, mercury, and cadmium. Improper handling and disposal of e-waste can result in significant soil and water contamination, as well as health risks for nearby communities.
GAI technologies can also indirectly contribute to environmental degradation through their applications. For example, AI-driven automation may lead to increased energy consumption in industries such as manufacturing, transportation, and agriculture. Furthermore, AI-powered systems may exacerbate resource extraction and consumption patterns by optimizing processes for efficiency without considering broader environmental implications.
The use of GAI therefore poses unique benefits and risks to climate change. On one hand, insights generated from predictive models can help scientists monitor the human progress of climate solutions. Deep learning applications can also track and predict metrics related to climate processes. But, GAI, and other forms of AI, can also potentially do more harm than good to the fight against climate change. By prioritizing sustainability and environmental stewardship, the AI community can attempt to mitigate the negative impacts of AI on the environment while also harnessing its transformative potential.
Treading Carefully into Uncertainty
GAI stands at the forefront of technological innovation, with a significant promise to revolutionize industries and reshape human interaction with technology. However, it's crucial to recognize and address the environmental costs associated with its development and utilization.
The energy-intensive nature of training and running LLMs like ChatGPT contributes significantly to carbon emissions, water consumption, and energy consumption, with data centers drawing from non-renewable sources. The reliance on fossil fuel energy exacerbates climate change, impacting ecosystems and biodiversity globally. Moreover, the production and maintenance of data center infrastructure entail substantial resource consumption and e-waste generation, further straining our environment.
As GAI permeates various sectors, its indirect impacts on energy consumption and resource utilization cannot be overlooked. While it holds promise in optimizing processes and aiding in climate monitoring, it also has the potential to exacerbate environmental degradation if not developed and utilized responsibly.
Amidst these challenges lies an opportunity for innovation and sustainability. By prioritizing energy efficiency, renewable energy adoption, responsible e-waste management, and sustainable resource sourcing, the AI community can attempt to mitigate the negative environmental impacts of GAI. Fostering interdisciplinary collaboration and ethical considerations in AI development can ensure that technological advancements align with environmental stewardship goals.
As we navigate the complexities of the digital age, we must carefully utilize the transformative potential of GAI while safeguarding the planet for future generations. By striking a balance between technological advancement and environmental responsibility, we can pave the way towards a more sustainable and equitable future.
Together, we must embark on a journey where innovation and environmental stewardship go hand in hand, ensuring that the use of GAI is directly in harmony with the health of our planet. Here are a few strategies that the AI community can leverage:
Strategies for Mitigating Environmental Impacts of GAI:
Energy Efficiency: Developing and implementing energy-efficient algorithms and hardware solutions is paramount to reducing the energy consumption of GAI models. Techniques such as model compression, quantization, and optimization may significantly decrease the computational requirements of training and inference processes, thereby lowering energy consumption and carbon emissions.
Renewable Energy Adoption: Transitioning data centers to renewable energy sources, such as solar, wind, and hydropower, is essential to minimize the environmental footprint of GAI. By investing in renewable energy infrastructure and power purchase agreements (PPAs), AI companies may reduce their reliance on fossil fuels and mitigate their contribution to climate change.
Responsible E-Waste Management: Implementing effective e-waste management practices is crucial to minimizing the environmental impact of outdated AI hardware. Recycling, refurbishing, and responsibly disposing of electronic equipment can help reduce the amount of electronic waste ending up in landfills and prevent the release of hazardous materials into the environment.
Sustainable Resource Sourcing: Ensuring sustainable sourcing of materials used in AI hardware production is vital for minimizing environmental degradation. Collaborating with suppliers to adhere to responsible mining practices, reducing material usage through design optimization, and promoting circular economy principles can help mitigate resource depletion and environmental pollution.
Ethical Considerations: Integrating ethical considerations into the development and deployment of GAI is essential for ensuring that technological advancements align with environmental stewardship and human rights goals. Prioritizing transparency, accountability, and fairness in AI algorithms and decision-making processes can help mitigate potential negative impacts on the environment and society.
While significant strides have been made in understanding and mitigating the environmental impacts of GAI, several challenges and opportunities lie ahead. One of the primary challenges is the need for interdisciplinary collaboration and knowledge exchange among AI researchers, environmental scientists, policymakers, and industry stakeholders. By fostering collective collaboration and sharing expertise across disciplines, we can develop holistic solutions that address the complex interplay between AI technology and environmental sustainability.
The rapid pace of technological innovation presents opportunities for the development of novel approaches and strategies to minimize the environmental footprint of GAI. Emerging technologies such as edge computing, federated learning, and neuromorphic computing hold some promise in reducing energy consumption, improving efficiency, and enabling decentralized AI infrastructure.
Raising awareness and promoting public engagement on the environmental impacts of GAI and AI development in general is essential for driving collective action and advocating for sustainable practices. Educating consumers, policymakers, and industry leaders about the importance of environmental stewardship in AI development and deployment can foster a culture of responsibility and accountability.
In Conclusion
In conclusion, addressing the environmental impacts of GAI requires concerted efforts from all stakeholders, including researchers, industry leaders, policymakers, and the public. By prioritizing energy efficiency, renewable energy adoption, responsible e-waste management, sustainable resource sourcing, and ethical considerations, we can attempt to mitigate the negative environmental impacts of GAI while utilizing its transformative potential.
As we navigate the path towards a more sustainable future, we must seize the opportunity to shape the trajectory of GAI development in harmony with environmental stewardship principles. Together, we can pave the way towards a future where technological innovation and environmental sustainability go hand in hand, ensuring a thriving planet for generations to come. As we fade away from obsolescence, let us create a new reality where we are conscious of mother Earth.
I leave you now with a set of questions: Can we use AI responsibly? What does AI’s future look like? In a society that is quick to adopt new technologies without first meditating on their broader societal impacts, it is of the utmost importance to be clear and knowledgeable on what technology is and is not. Consider this impactful quote from Hubert Dreyfus:
“During the past two thousand years the importance of objectivity; the belief that actions are governed by fixed values; the notion that skills can be formalized; and in general that one can have a theory of practical activity, have gradually exerted their influence in psychology and in social science. People have begun to think of themselves as objects able to fit into the inflexible calculations of disembodied machines: machines for which the human-form-of-life must be analyzed into meaningless facts, rather than a field of concern organized by sensory-motor skills. Our risk is not the advent of superintelligent computers, but of subintelligent human beings.” (What Computers Still Can’t Do)
~ Marco Antonio Lueras, New Mexico Conservation Alliance
Author’s Bio
Marco Antonio Lueras is the Co-Founder and Executive Director of New Mexico Conservation Alliance, a non-profit based in the Southwestern U.S. with the mission of empowering New Mexico's youth to become stewards of their environment by connecting them with meaningful service opportunities in collaboration with water and environmental organizations. An avid research pioneer, Marco serves on several advisory boards for companies working in the intersections of AI and water. As a scholar, Marco is currently working towards a PhD from the University of Miami that critically examines the human-technology relationship in criminal justice contexts. Marco is passionate about environmental conservation, human rights, and the pursuit of knowledge.
If you want to contact the author please visit his LinkedIn here: https://www.linkedin.com/in/marcolueras/
I have been waiting for a long time for someone to write an in-depth analysis of this topic, which is as interesting as it is necessary. The impacts of AI are not only in the digital, but also in the physical world, and we need to take this into account in discussions on related topics.