There’s a hum in the warm summer air in Lowell, Massachusetts. A low hum, more felt than heard. What could it be? Cicadas? A small plane overhead? Perhaps something much larger, more permanent, and more concerning to public health: an AI data center.

During the summer after my first year of medical school, I was asked by everyone back home my thoughts on AI in medicine. In those conversations, I saw a glaring gap in the discussed dangers of AI, among the general public and medical professionals alike. Whether you are pro-AI, anti-AI, or believe that either label quashes necessary nuance, we have all heard alarms being raised about AI and bias magnification, hallucinations, intellectual property ethics, security concerns, and environmental damage. The last of these is not just a concern for environmentalists. It is a public health crisis waiting to happen and sliding under the radar.

There are three main public health concerns when it comes to AI: air pollution, water usage, and noise pollution. AI data centers use a lot of energy, drawing as much in a year as 100,000 households do, four to 10 times more than conventional data centers. This energy is often produced by gas turbines, with diesel generators as a backup. Both of these expel fine particulate matter into the air. This matter, an IARC group 1 carcinogen, contributes to respiratory, cardiovascular, neurodegenerative, and reproductive disease, making it a leading environmental contributor to early death.

Often discussed is AI’s use of water. Large amounts of water are needed for creating and cooling data centers. Centers built in water-stressed regions, of which there are many, threaten drinkable water supplies. Dehydration is already a leading cause of infant mortality in South Asia and sub-Saharan Africa. Furthermore, when access to water is limited, sanitation suffers, skyrocketing the rates of water-borne diseases like cholera.

The least studied health concern of AI is noise pollution. It doesn’t create the noise we may be used to, freeway traffic and construction. Rather, data centers generate low-frequency infrasound that doesn’t register on the standard decibel scale, meaning it goes largely unregulated by city zoning laws. Residents near data centers, such as those in New Jersey and Michigan, report sleep deprivation, headaches, and anxiety. When sleep is so important for cognitive function and immune function and long-term reduction of cardiovascular disease, dementia, and depression risk, it is easy to see how 24/7 exposure to infrasound can become a major detriment to health.

It’s crucial to note that these health effects are not distributed equally. Water-stressed regions, particularly impoverished regions, bear the brunt of water crises. Air pollution already disproportionately impacts the Global South. Data centers have great potential to further widen that gap. Noise pollution affects the communities where data centers are built, poorer and more rural communities that have less political leverage to fight construction in their neighborhood. Yes, AI is bringing exciting new opportunities to medicine, but it’s important to ask ourselves if it is creating solutions that will be mainly accessible to the privileged at the expense of the health of the most vulnerable.

As troubling as this is, AI-driven global health crises are not a foregone conclusion. The health impacts of AI are highly dependent on where and how a data center is operated. Centers in cooler, water-abundant climates create less water stress. Centers powered by clean energy would put less particulate matter into the air. And centers built away from homes are less likely to cause noise-related disturbances to people, though infrasound may have detrimental effects on wildlife as well.

There are studies that show that in the future, AI could help slow environmental destruction. However, this is only if the aforementioned damage mitigation strategies are taken, but these won’t be implemented on their own. Strong governmental regulation and incentivization is necessary because these measures are a far cry from how data centers are currently operating. They require a time and cost investment companies are not keen to take on unless required to. AI technology is outpacing regulations, and many countries do not currently seem interested in enacting the legislation needed to make it an environmental net positive.

It’s here that health care professionals have a role to play. Despite the hurdles of recent years, physicians still enjoy some social sway on health issues. Taking AI’s impact on public health seriously puts pressure on legislators. Physicians are often hesitant to get involved in politics, with physician voting rates being 12 percent less than non-physicians in the U.S. But as we see more and more, policy, environment, and health are intertwined, and it is our responsibility to pay attention to these factors that have an impact on our patients’ health.

Beyond pushing institutions and governments to do their regulatory part, there are individual efforts that can be made too. Data centers are only constructed because companies believe in the demand. As common as AI is now, we do not have to always use it. We didn’t have AI chatbots five years ago. Before you fire up ChatGPT, Gemini, Copilot, or Claude, take a moment to ask yourself how much you really need it. A non-AI search engine can look up restaurants near you or tell you when the DSM-5 was revised just as well. And if you do decide you need it, you can still lower your resource usage. Decreasing the number of prompts used will decrease energy used. For example, being “polite” to AI, saying “please” and “thank you,” eats up a significant amount of energy. These days, many apps and websites have AI features built in. Many of these can be turned off. If you don’t actually need AI to haphazardly summarize an email for you or give you syntax suggestions when writing a paragraph, and it’s likely you don’t, then taking a second to disable these features will save resources. AI may feel ubiquitous, but we don’t need to blindly accept that it and its environmental costs are a given, especially not when the health and well-being of our patients hangs in the balance.

Jessie Hsu is a medical student.