A study conducted by the California Institute of Technology (Caltech) and the University of California Riverside (University of California Riverside) found that the incredible power demands of AI will cause dangerous air pollution. This will lead to the appearance of asthma symptoms in 600 thousand people and 1.3 thousand premature deaths annually, as well as an increase in health care costs by $20 billion/year by 2030, reports eWeek.
The Unpaid Toll: Quantifying the Public Health Impact of AI examines the disruptive impact of data center AI on society. The more data centers and related infrastructure appear, the more energy is required to power them and the higher the emissions of power plants, as well as backup diesel generators. At the same time, emissions contain microparticles less than 2.5 microns in size (PM2.5), sulfur dioxide (SO2) and nitrogen dioxide (NO2), which negatively affect humans.
According to research results, low-income citizens living near power plants and data centers are most at risk from rising diseases. Thus, backup diesel generators emit 200–600 times more nitrogen dioxide compared to gas power plants, but the matter is not limited to pollution of nearby areas. The emissions also poison neighboring states, affecting even areas far from the original point of pollution.
Image source: Janusz Walczak/unsplash.com
For example, pollution from diesel generators in a data center in Northern Virginia, which serves as the main data center hub in the United States and perhaps the world, also extends to West Virginia, Maryland, Pennsylvania, New York, New Jersey, Delaware, and even reaches Washington. Even if generator emissions are kept within 10% of acceptable limits, regional health care costs would be $190 million to $260 million annually. If generator emissions are at the maximum permissible level, then costs will increase by an order of magnitude.
The work highlights the scale of the problem. Training one Llama 3.1 large language model (LLM) produces emissions equivalent to 10,000 trips in a regular car between Los Angeles and New York. By 2030, AI-driven pollution could rival the emissions of 35 million vehicles. AI will even surpass the steel industry in its impact on public health.
Image source: UC Riverside/Caltech
Scholars highlight the lack of corporate responsibility. According to them, if you look at the sustainability reports of technology companies, you will notice that they contain information mainly about carbon emissions and water consumption, which are already creating a burden on healthcare (previously it was reported that in just one session a bot like ChatGPT “drank” about 0.5 liters of water). There is no mention at all of the pollution that is already straining the health care system.
The study’s authors call on AI companies to report more transparently, not just on carbon emissions, but also on other air pollutants. It is also recommended to provide compensation to affected communities and implement sustainable technologies to minimize environmental damage.
Meanwhile, energy shortages in the world and the United States due to AI data centers are only growing, and air pollution is not the only problem. It turned out that data centers upset the balance in the energy supply system of residential buildings in the United States, which in the worst case can even lead to fires in houses near the data center. By order of the current US President, federal lands will be allocated for the construction of AI infrastructure and green energy facilities.