AI’s Hidden Threat to Public Health

Most people have heard about the environmental impact of the mine of artificial intelligence today, caused by the sprawling data centers full of energy -thirsty servers. In the United States alone, the demand for artificial intelligence is expected to pay electricity consumption at the data center 6.7 to 12.0 percent of the country’s total by 2028. By the same date, water consumption is expected to double to cool this or even quadruple data center facilities, compared to 2023.

But many people were not associated with data centers and public health. The power plants and backup generators needed to maintain data centers that work by generating harmful air pollutants, such as the microscopic substance and nitrogen oxides (NOX). These pollutants affect human health, which leads to symptoms of asthma, heart attacks, and even cognitive decline.

But the contribution of artificial intelligence to air pollution and the burden of public health is often missing from talks about responsible artificial intelligence design. Why?

Because the surrounding air pollution is “A silent killer.” While concerns about General health effects of data centersIncluding possible links to increase the cancer rate, began on the surface, most of the male model developers, practitioners and users simply do not realize the serious health risks associated with energy and infrastructure that operate modern artificial intelligence systems.

Hanger pollution of the surrounding air

The surrounding air pollution is responsible for Nearly 4 million early deaths Throughout the world every year. The largest of the perpetrator is the small particles of 2.5 ور m or less in its diameter (referred to as the name PM 2.5), Which can be deeply transmitted in the respiratory system and lungs. Along with high blood pressure, smoking, and high blood sugar, air pollution is a pioneering health risk factor. The World Bank estimates The global cost of air pollution at 8.1 trillion USDThat is, 6.1 percent of Global GDP.

Unlike the common belief, air pollutants do not remain near their emissions: they can travel hundreds of miles away. Moreover, PM 2.5 contaminated is “non -threshold”This means that there is no safe level of exposure.

With the danger of this pollution firmly, the question becomes: How much is it responsible for artificial intelligence? in We searched Caltech professors and the University of California, Riferside, we started answering this question.

Measuring the general health cost of Amnesty International

To ensure the availability of artificial intelligence services even during network interruptions, data centers depend on large groups of reserve generators that usually burn diesel fuel. While the total operating time for backup generators is limited and organized by local environmental agencies, their emissions rates are high. A model diesel generator can launch 200 to 600 times more than nitrogen oxides From the natural gas power plant produces the same amount of electricity.

accident a report Virginia has revealed that backup generators in Virginia data centers emit about 7 percent of permissible permits in 2023. According to the US Environmental Protection Agency Cobra modeling toolWhich draws how air pollution affects human health at local, state and federal levels, the cost of the general health of these emissions in Virginia is estimated at $ 150 million, which affects distant societies such as Florida. Imagine the effect if the data centers have reached the maximum permitted emissions.

What increases the increase in public health risks, a large group of data center generators in the region may work simultaneously during the interruption of the network or a lack of networks as part of the response programs, which may lead Harmful for people with lung problems.

Next, let’s look beyond the reserve generators to supply energy from the network. The largest part of the data centers that operate the electricity comes from power plants that burn fossil fuels, which launch harmful air pollutants, including PM 2.5 and Nox. Despite the years of progress, the power plants remain The main source of air pollution In the United States.

We calculated that the great Tructured IQ in the United States, such as Meta’s Llama 3.1, could result in more than PM 2.5 10,000 round trips in the car Between Los Angeles and New York City.

According to our research, in 2023, the air pollution of the lineage to the American data centers was responsible for It is estimated at 6 billion dollars in public health damage. If the current development of the current artificial intelligence continues, this number is expected to reach $ 10 billion to $ 20 billion annually by 2030, which competes with the impact of emissions of 30 million cars in California.

Why carbon and energy energy are not the entire story

To date, the efforts made to relieve the environmental footprint of artificial intelligence mostly focused on carbon emissions and energy efficiency. These efforts are important, but they may not relieve the health effects, which are highly dependent on the place where emissions occur.

Carbon anywhere is carbon everywhere. The climate effect of carbon dioxide is largely the same regardless of where it is emitted. But the health effect of air pollution depends heavily on regional factors such as local sources of energy, wind patterns, weather and population density.

Although carbon emissions and harmful harmful air pollutants have some common sources, the exclusive focus on carbon cut does not necessarily reduce, and may exacerbate the risk of public health. For example, our recent (and unpublished) research showed that the redistribution of Meta energy loads in 2023 through American data centers to determine the priorities of carbon discounts may reduce the total carbon emissions by 7.2 percent, but will increase public health costs by 2.8 percent.

Likewise, the focus only on energy efficiency can reduce air pollutants emissions, but does not ensure a low health effect. This is because training the artificial intelligence model itself using the same amount of energy can result in significantly different health results depending on the site. Through the American data centers in Meta, we found that the general health cost of training the same model could vary in more than 10 factor.

We need the authenticity of artificial intelligence

The supply side solutions, such as the use of alternative fuel for backup generators and electricity design from clean fuel, can reduce the impact of the general health of Amnesty International, but they come with great challenges.

Clean backup generators that offer the same level of diesel reliability are still limited. Although progress in renewable energy, fossil fuels remain in depth in the mixture of energy fuel. US Energy Information Management Projects In 2050, the coal -based electricity generation will remain about 30 percent of the level of 2024 under the alternative electricity scenario, as power plants continue to work under the rules before April 2024. I remained almost flat Over the past four decades, it confirms the difficulty of changing energy supplies that operate the entire data centers.

We believe that the demand strategies that consider spatial and temporal differences in health effects can provide effective and implemented solutions immediately. These strategies are especially suitable for artificial intelligence data centers with great operational flexibility. For example, artificial intelligence training can be operated in any available data centers and usually does not face difficult final dates, so that these jobs can be directed to sites or postponed to times that have a lower impact on public health. Likewise, inference functions – the work of the model to create outputs – can be directed between multiple data centers without affecting the user experience.

By integrating the effect of public health as a major performance of performance, this flexible citizen can be harnessed to reduce the growing burden of artificial intelligence. It is important, this persuasive approach to health in artificial intelligence requires minimal changes to current systems. Companies simply need to consider public health costs when making decisions.

While Amnesty International’s public health is growing rapidly, artificial intelligence carries a tremendous promise to promote public health. For example, within the energy sector, Amnesty International can move in the complex decision space to send the power plant in actual time. By align the stability of the network with public health targets, artificial intelligence can help reduce health costs while maintaining reliable energy supplies.

Artificial intelligence has become at a general benefit and will continue to reshape society deeply. Therefore, we must examine artificial intelligence through a general lens, with its impact on public health as a critical consideration. If we continue to overlook it, the general cost of Amnesty International will only grow. Affective health intelligence provides a clear way forward to develop artificial intelligence while strengthening cleaner air and health societies.