AI’s hidden threat to public health

Most people have heard about the environmental impact of today’s AI, derived from extended data centers full of hungry servers of energy. Only in the United States, it is projected that the demand for impulse the consumption of electricity from the data center to 6.7 to 12.0 percent of the total nation for 2028. On that same date, the water consumption to cool these data centers facilities is predicted that will double, or even quadruple, compared to level 2023.

But many people have not established the connection between data centers and public health. Electrical plants and backup generators needed to maintain data centers that work generate harmful air pollutants, such as fine particles and nitrogen oxides (NOX). These pollutants have an immediate cost in human health, triggering asthma symptoms, heart attacks and even cognitive impairment.

But AI’s contribution to air pollution and the public health burden often lack in conversations about the design responsible for AI. Because?

Because ambient air pollution is a “Silent murderer.” While concerns about the Public health impacts of data centersincluding possible links to Cancer rate increasesThe majority of developers, professionals and users of the AI ​​model are simply not aware of the serious health risks linked to energy and infrastructure that modern the systems.

The danger of ambient air pollution

Environmental air pollution is responsible for Approximately 4 million premature deaths worldwide every year. The greatest culprit is the small particles 2.5 micrometers or less in diameter (called PM 2.5), which can travel deeply in the respiratory tract and lungs. Together with high blood pressure, smoking and the high level of blood sugar, air pollution is a leading health risk factor. The World Bank estimates the Global cost of air pollution at US $ 8.1 billionequivalent to 6.1 percent of the global gross domestic product.

Contrary to common belief, air pollutants do not remain close to their emission sources: hundreds of miles can travel. Besides, PM 2.5 is considered a “non -threshold” pollutantwhich means there is no safe exposure level.

With the danger of this well established pollution, the question is: how much is it responsible for AI? In Our research As professors of Caltech and the University of California, Riverside, we have set out to answer that question.

Quantification of the public health cost of AI

To ensure that IA services are available even during grid interruptions, data centers depend on large sets of backup generators that generally burn diesel fuel. Although the total operating time of support generators is limited and regulated by local environmental agencies, their issuance rates are high. A typical diesel generator can be released 200 to 600 times more nox than a natural gas energy plant that produces the same amount of electricity.

A recent one report For the state of Virginia, he revealed that support generators in the Virginia data centers issued around 7 percent of what the permits allowed in 2023. According to the United States Environmental Protection Agency. Cobra Modeling ToolThat maps how air pollution affects human health at the local, state and federal level, the cost of public health of these emissions in Virginia is estimated at $ 150 million, which affects communities as far as Florida. Imagine the impact if data centers reach their allowed emissions.

More than the public health risk increases, a large set of data centers generators in a region can operate simultaneously during networks of networks or network shortage as part of demand response programs, which potentially triggers short -term peaks in PM2.5 and NOx emissions that are especially. harmful to people with lung problems.

Next, let’s see beyond the support generators to the supply of energy from the network. Most of the electricity food data centers come from energy centers that burn fossil fuels, which release harmful air pollutants, including PM 2.5 and NOX. Despite the years of progress, electric plants are still Leading air pollution source In the United States.

We calculate that the training of a single large generative model in the United States, as goal is called 3.1, can produce both PM 2.5 and more than 10,000 round trips by car Between Los Angeles and New York City.

According Our researchIn 2023, air pollution attributed to US data centers Estimated $ 6 billion in damages to public health. If the current growth trend of the AI ​​continues, it is projected that this number from $ 10 billion to $ 20 billion per year by 2030, rivaling the impact of the emissions of the 30 million vehicles in California.

Why carbon and energy efficiency are not the whole history

To date, efforts to mitigate the environmental footprint of AI have focused mainly on carbon emissions and energy efficiency. These efforts are important, but may not relieve health impacts, which depend largely where emissions occur.

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

Although carbon emissions and air pollutants that damage health have some shared sources, an exclusive approach in cutting carbon is not necessarily reduced and could even exacerbate public health risks. For example, our latest research (and unpublished) has shown that redistributing target energy loads in 2023 in their US data centers to prioritize carbon reductions could potentially reduce carbon emissions by 7.2 percent, but increase public health costs by 2.8 percent.

In the same way, focusing solely on energy efficiency can reduce emissions of air pollutants, but does not guarantee a decrease in health impact. This is because they train the same AI model that uses the same amount of energy can produce very different health results according to the location. In all finishing data centers in the United States, we have discovered that the cost of public health of the training of the same model can vary in more than one factor of 10.

We need health informed

The solutions on the supply side, such as the use of alternative fuels for support generators and electricity of clean fuels, can reduce the impact of AI public health, but come with significant challenges.

Clean support generators that offer the same level of reliability as diesel are still limited. And despite the advances in renewable energy, fossil fuels remain deeply integrated in the combination of energy fuel. The United States Energy Information Administration projects This generation of coal -based electricity in 2050 will remain at approximately 30 percent of the 2024 level under the alternative electricity scenario, in which electrical plants continue to operate under the existing rules before April 2024. It remained almost flat In the last four decades, underlining the difficulty of completely changing the energy supply that feeds the data centers.

We believe that the strategies of the demand side that the spatial and temporal variations in health impacts consider can provide effective and processable solutions immediately. These strategies are particularly suitable for AI data centers with substantial operational flexibility. For example, the training of AI can often be executed in available data centers and, usually, do not face difficult deadlines, so those works can be enrupted locations or deferred in times that have less impact on public health. Similarly, inference work, the work that does a model to create a way out, can be enruted among multiple data centers without affecting the user experience.

By incorporating the impact on public health as a key performance metric, these flexibilities can be used to reduce the growing health burden of AI. Crucially, this approach informed by health for AI requires minimal changes in existing systems. Companies simply need to consider public health costs when making decisions.

While the public health cost of AI is growing rapidly, AI also has a great promise to advance public health. For example, within the energy sector, AI can navigate the complex decision space of the energy plant office in real time. By aligning network stability with public health objectives, AI can help minimize health costs while maintaining a reliable energy supply.

The AI ​​is quickly becoming a public utility and will continue to remodel society. Therefore, we must examine AI through a public lens, with its impact on public health as a critical consideration. If we continue to overlook it, the cost of public health of AI will only grow. The AI ​​informed in health offers a clear path to advance in AI while promoting cleaner air and healthier communities.