Air Pollution And Ai: A Toxic Relationship

AI has the potential to improve societal systems and the way we work and play. However, the environmental impact of AI is a growing concern. AI is housed in data centres, which consume vast amounts of electricity, water, and rare earth elements, and produce electronic waste. AI's electricity consumption is a significant contributor to air pollution, with the generation of electricity for AI data centres expected to cause up to 1,300 premature deaths per year in the US by 2030. The health costs of AI-related air pollution are estimated to reach $20 billion within six years. The impact of AI on the environment is a critical issue that needs to be addressed to ensure a sustainable future.

AI electricity consumption and its impact on asthma

The electricity consumption of AI is expected to grow significantly in the coming years. In 2023, AI represented a power consumption of 4.5 gigawatts worldwide, or about 8% of total power consumption in data centres. By 2028, AI's power consumption is projected to reach 14 to 18.7 gigawatts, accounting for up to 20% of data centre power consumption. This surge in electricity demand is driven by the increasing number of data centres, which are expected to more than double by 2030, with AI being the most significant driver of this increase.

The growing electricity consumption of AI has direct implications for public health, particularly in the context of air pollution and its impact on asthma. Data centres, which house AI servers, rely on power plants and backup generators that emit harmful pollutants such as fine particulate matter and nitrogen oxides (NOx). These pollutants are known to trigger asthma symptoms and can have immediate and tangible effects on human health.

The impact of AI's electricity consumption on asthma is already being felt, with researchers estimating that the gas-powered generators used as backups in Virginia's Data Center Alley could be contributing to 14,000 cases of asthma symptoms, resulting in significant public health costs. By 2030, it is estimated that the electricity consumption of AI could contribute to approximately 600,000 asthma cases in the United States alone. This figure represents more than a third of annual asthma deaths in the country.

The public health burden of AI's electricity consumption is not limited to asthma but also includes other respiratory and cardiovascular issues, as well as potential cognitive decline. The emissions from data centres can affect communities far beyond their immediate vicinity, as pollutants can drift across county and state lines. This widespread impact underscores the urgency of addressing the hidden costs of AI's electricity consumption on public health, particularly for vulnerable communities, including those with pre-existing health conditions.

AI's contribution to air pollution and its health costs

AI is expected to become a significant part of our lives, offering clear benefits that have the potential to improve societal systems. However, it is important to address the negative impacts of AI, such as its contribution to air pollution and the resulting health costs.

AI's contribution to air pollution is often overlooked, as most people don't associate data centers with public health risks. The power plants and backup generators needed to keep data centers running generate harmful air pollutants, including fine particulate matter and nitrogen oxides (NOx). These pollutants have immediate and severe impacts on human health, including triggering asthma symptoms, heart attacks, and cognitive decline. They are also linked to potential cancer rate increases.

The growth of AI is driving a significant increase in the demand for data centers and energy, making it the fastest-growing sector for energy consumption across all industries. This increase in energy consumption leads to a steep rise in air pollution. The public health burden of AI-related air pollution is projected to be double that of the U.S. steel-making industry by 2030 and could rival the emissions from all the cars, buses, and trucks in California.

The health costs of AI-related air pollution are significant. In the United States alone, air pollution attributed to data centers was responsible for an estimated $6 billion in public health damages in 2023. If the current AI growth trend continues, this number is expected to reach $10 billion to $20 billion per year by 2030. Additionally, the pollution from data centers affects communities far beyond their immediate vicinity, as pollutants drift across county and state lines. For example, pollution from backup generators at data centers in Northern Virginia impacts multiple states, resulting in regional public health costs of $190 million to $260 million annually, which could increase tenfold if emissions reach their maximum permitted level.

To address these issues, researchers recommend adopting standards and methods that require tech companies to report the air pollution caused by their power consumption and backup generators. They also suggest that tech companies compensate communities disproportionately affected by air pollution, particularly low-income communities located near power plants or data centers.

The environmental impact of AI training

AI training's environmental impact is twofold: the direct effect of energy consumption and the indirect consequences stemming from the proliferation of data centres. The former refers to the substantial energy demands of training and running AI models, contributing to a staggering carbon footprint. The latter pertains to the environmental toll of the physical infrastructure housing AI servers, encompassing electronic waste, water consumption, unsustainable mining practices, and greenhouse gas emissions.

AI training's direct environmental impact lies in its energy intensity. The computational requirements for training AI models have skyrocketed, with computing power doubling every 3.4 months since 2012. This surge in computational needs translates to a substantial carbon footprint. For instance, training large AI models can generate approximately 626,000 pounds of carbon dioxide, equivalent to around 300 round-trip flights between New York and San Francisco. This increase in energy use exacerbates greenhouse gas emissions, contributing to climate change.

The indirect environmental consequences of AI training are equally concerning and stem from the proliferation of data centres. These centres, housing AI servers, produce electronic waste, consume substantial water resources, rely on unsustainable mining practices for rare elements, and contribute to planet-warming greenhouse gas emissions. The electronics housed in data centres require a significant amount of raw materials, and the construction and operation of these facilities further contribute to their environmental footprint.

To address these challenges, a multifaceted approach is necessary. Standardized procedures for measuring AI's environmental impact and greater transparency in the creation and application of AI systems can help. Additionally, investing in energy-efficient hardware and algorithms can significantly reduce energy consumption without compromising AI performance.

The use of water in AI data centres

AI data centres are major consumers of water, and their water usage is a growing concern. AI servers have extremely high energy consumption, which generates a lot of heat. To prevent overheating, data centres use water-based cooling systems, which require a significant amount of clean, fresh water. This water consumption is referred to as "scope-1" onsite water usage.

The water is used in cooling towers, which rely on water evaporation to produce cold water. Additionally, outside air needs water for evaporation assistance when temperatures are high and for humidity control in dry conditions. This process of heat dissipation ensures that the servers remain within safe operating temperatures, preventing system failures, data corruption, and costly downtime.

The second type of water consumption, "scope-2", is related to offsite electricity generation. The high power densities of AI servers require a lot of electricity, which is often generated at thermal power, nuclear, and hydropower plants. These power plants consume a significant amount of water through their cooling processes, and in the case of hydropower plants, expedited water evaporation.

The water footprint of AI data centres is a critical issue, especially in regions prone to droughts and water scarcity. As the AI industry continues to grow, it is important to address this issue and reduce water consumption. Some companies have started taking steps towards water conservation, aiming to achieve "water positive" goals. Additionally, forward-thinking data centres are adopting water reuse and recycling technologies to reduce their water usage and environmental impact.

AI's impact on the planet and the need for regulation

AI has an immense environmental impact, and its rapid growth has led to a significant increase in energy and water use, as well as carbon emissions. AI technologies, such as chatbots, have been scrutinized for their high energy demands, which often come from burning fossil fuels, the biggest contributor to global warming. The production, transport, maintenance, and disposal of AI hardware devices also require substantial materials and natural resources.

The proliferating data centers that house AI servers produce electronic waste and are large consumers of water, a resource that is becoming increasingly scarce. The microchips that power AI need rare earth elements, which are often mined in environmentally destructive ways. Data centers also use water during construction and to cool electrical components once operational. The number of data centers has surged to 8 million from 500,000 in 2012, and experts expect AI's demands on the planet to keep growing.

The environmental impacts of AI are not limited to physical resources but also extend to social and political realms. For instance, when asked about environmental challenges, chatbots reflected biases and avoided mentioning radical solutions, such as "environmental justice" and "dismantling colonialism." This has implications for how we understand and communicate climate change and its solutions.

Given the far-reaching consequences of AI on the planet, there is a pressing need for regulation and sustainable practices. The United Nations Environment Programme (UNEP) has made recommendations to address AI's environmental footprint, including establishing standardized procedures for measuring AI's environmental impact and developing regulations for companies to disclose the direct environmental consequences of AI-based products and services. Countries can also encourage the use of renewable energy and carbon emissions offsetting in data centers. Through proactive measures and ethical practices, it is possible to strive for a future where AI and environmental preservation go hand in hand.

Frequently asked questions