AI’s rapid growth fueling public health crisis

The findings of this study serve as a stark reminder that the health implications of AI cannot be ignored. It calls for a new, health-conscious approach to AI development and operation, one that takes into account the public health costs and aims to mitigate its harmful effects on vulnerable communities.

The rapid expansion of artificial intelligence is quietly contributing to a public health crisis. While much attention has been given to AI's environmental footprint, its effects on air quality and public health are largely ignored. A recent study by researchers at the University of California, Riverside (UCR) and the California Institute of Technology (Caltech) highlights the human impacts associated with AI's energy consumption and also offers urgent recommendations for mitigating these harms.

The study, titled "The Unpaid Toll: Quantifying the Public Health Impact of AI," is published online in the pre-print journal arXiv. The paper authors include Shaolei Ren, a UCR associate professor of electrical and computer engineering, Caltech professor and computer scientist Adam Wierman, and Yuelin Han, Zhifeng Wu, and Pengfei Li, all three with UCR’s Bourn’s College of Engineering. 

The hidden toll of AI

The expansion of AI and its reliance on energy-intensive data centers are driven by the massive computational requirements of modern AI models. As the demand for generative AI and large-scale machine learning models increases, data centers are proliferating rapidly. These centers consume vast amounts of electricity, much of which is still sourced from fossil fuels, contributing not only to carbon emissions but also to harmful air pollutants that degrade air quality.

The lifecycle of AI -from hardware manufacturing to the operation of data centers - directly contributes to the release of several types of "criteria air pollutants," including PM2.5 (fine particulate matter), sulfur dioxide (SO2), and nitrogen dioxide (NO2). These pollutants are linked to a variety of adverse health outcomes, such as premature mortality, asthma, heart attacks, strokes, lung cancer, and cognitive decline, particularly among vulnerable groups such as the elderly, children, and those with pre-existing respiratory or cardiovascular conditions.

The research highlights that AI's environmental costs are not confined to greenhouse gas emissions alone. While carbon emissions from energy consumption contribute to long-term climate change and have broader, indirect health implications, the immediate threat posed by air pollutants like PM2.5 is far more direct. In fact, PM2.5 exposure accounts for approximately 1 million premature deaths globally each year and represents 2% of all deaths worldwide from 2000 to 2019. Even short-term exposure to PM2.5, which can be generated by AI operations, is linked to severe health consequences, including death.

A growing health burden

The public health costs associated with AI's energy consumption are staggering. According to the study, by 2030, the health burden resulting from U.S. data centers is projected to exceed $20 billion annually—this amount is more than double the public health cost of U.S. coal-based steelmaking, which is a significant source of air pollution. In fact, the public health impact of AI-driven pollutants could rival or even exceed the cost of air pollution from on-road emissions in the largest U.S. states, including California, with approximately 35 million registered vehicles.

To put this in perspective, the emissions generated by training AI models like Llama-3.1 can release an amount of air pollutants equivalent to more than 10,000 round trips between Los Angeles and New York City by car. The health-related economic costs associated with these emissions, such as hospitalizations, medication usage, school loss days, and work absences, can far exceed the cost of the electricity used to power AI systems. This stark contrast underlines the urgent need to address AI’s direct and immediate health impacts alongside its environmental footprint.

The study also emphasizes that the public health burden is unevenly distributed across different regions. Economically disadvantaged communities are disproportionately affected by the health impacts of AI data centers, where per-household health costs can be 200 times higher than those in less-impacted areas. In communities like Meigs County, Ohio, the health burden can equate to nearly eight months of household electricity bills, further exacerbating social inequities. As the demand for AI continues to grow, this disparity is expected to increase, particularly in communities that host or rely on data centers but do not benefit directly from the economic gains associated with AI operations, such as tax revenue.

Disproportionate impact

One of the most alarming findings of the study is the disproportionate health impact on low-income communities, which are often the most vulnerable to the effects of air pollution. The study shows that in 2030, economically disadvantaged regions will bear a health cost more than 200 times greater than wealthier areas. This disparity in health outcomes is a direct result of the placement of data centers in or near such communities, where they often lack the resources to mitigate these impacts. These communities also tend to have lower access to healthcare and fewer means to combat the health effects of increased air pollution.

The public health burden in these areas is further compounded by the transportation of pollutants over large distances. Pollutants emitted by AI data centers, especially from fossil-fuel-powered backup generators, can travel hundreds of miles, impacting communities far beyond the immediate vicinity of the data centers themselves. The study shows that this cross-state pollution is particularly harmful to those in regions already struggling with poor air quality.

Recommendations 

Given the significant and growing public health costs associated with AI’s energy consumption, the study offers several key recommendations for mitigating these harms:

• Standardized Reporting Protocols for Pollutants: The study calls for the adoption of standardized reporting protocols for criteria air pollutants and their associated public health costs. Just as the tech industry has made strides in reporting carbon emissions through frameworks like the greenhouse gas protocol, a similar system should be developed to track and report the health impacts of air pollution generated by AI’s energy consumption. Transparency in this regard is essential for understanding and addressing the public health burdens of AI.

• Health-Informed AI Operations: To reduce the public health impact of AI, the study recommends the adoption of health-informed AI. This approach would involve optimizing the scheduling of AI model training and inference tasks based on regional health data. By exploiting the flexibility in data center operations and taking into account the varying health costs associated with electricity generation in different regions, AI operations can be scheduled to minimize their health impacts. For example, AI tasks could be shifted to regions with lower air pollution or periods of lower emissions to reduce the overall health burden.

• Promoting Public Health Equity: The study stresses the importance of promoting public health equity by ensuring that the most vulnerable communities, particularly low-income areas, are not bearing an unfair share of the public health burden. This can be achieved by prioritizing policies that reduce the health impacts in these communities and ensuring that AI-related industries contribute to improving health outcomes in the regions they affect. Policymakers should consider the long-term health impacts of AI when deciding where to site data centers and how to allocate resources to mitigate the associated health risks.

A call for responsible AI development

The findings of this study serve as a stark reminder that the health implications of AI cannot be ignored. It calls for a new, health-conscious approach to AI development and operation, one that takes into account the public health costs and aims to mitigate its harmful effects on vulnerable communities.

With the right policies in place, AI can indeed help improve public health, but its development must be approached with a careful consideration of its broader social and health impacts. The growth of AI should not come at the cost of human health, and the tech industry must take responsibility for its role in shaping a healthier, more equitable future.