From Fax Machines to Forecasts: What Artificial Intelligence Can Already Do for Public Health Practice

When people imagine AI in medicine, they picture a statistical model reading a x-ray or suggesting a diagnosis for a single patient. Public health AI works on a different object entirely. It is built to improve decisions about populations, communities, and systems, and it is judged by a different standard: not diagnostic accuracy for one person, but whether prevention is better targeted, outbreaks are caught earlier, inequities shrink, and a stretched workforce can do more. Having spent much of my career inside surveillance and outbreak systems at the local, national, and international level, I can describe concretely what the technology can already do across four domains of practice.

Disease surveillance and early warning

The foundation of public health is surveillance: the system by which governments count events in a population, from salmonella infections to suicides to lead poisoning, and use those counts to detect outbreaks, evaluate programs, and direct resources. In much of the United States this work still runs on human labor and physical paperwork, with reports arriving by fax or telephone, demographic fields left blank, and staff placing phone calls to fill the gaps. Surveillance is fundamentally a problem of moving, cleaning, and categorizing large volumes of data quickly, which is exactly the problem AI has already transformed in finance and logistics. Natural language processing can read a free-text laboratory report and extract the pathogen, the test, the date, and the reportable patient details; it can recognize a reportable disease as it streams off an instrument and trigger a report without waiting for a person to notice; and it can match records that refer to the same patient despite different spellings. It can also run the nowcasts and forecasts that today only well-resourced health departments can produce, and that most county departments cannot produce at all. The cost of not counting is real, as I have written about the way the United States does not know how many people have alpha-gal syndrome, and that ignorance is itself the problem.

Emergency and outbreak response

When an agency declares an emergency, it activates an incident management structure and reassigns staff into roles for leadership, logistics, operations, and communications, and the people in those roles are routinely overwhelmed by the volume of information and decisions a response demands. AI can improve emergency response in several practical ways. During a large response, meetings may run every eight hours, every day, each requiring rapid synthesis and careful documentation for both coordination and the legal record. Ambient AI can serve as a real-time recorder of decisions and open questions and can draft the situation reports that leadership and partner agencies require, while a system built into the response from the start can track the timeline as it happens rather than reconstructing it months later. The same capability supports the logistics of countermeasures, monitoring inventory across distribution points, modeling future need against the projected course of an outbreak, and supporting registration, eligibility, and adverse-event detection at the sites that dispense them. These capabilities apply equally to overdoses, unintentional injuries, and rising infant mortality, conditions that rarely command the attention an infectious outbreak does.

Preventing disease at population scale

The largest and most expensive burden of disease in this country comes from chronic conditions that develop over years and generate enormous ongoing costs once established. Clinicians are asked to help patients prevent them, but not everyone has access to primary care, and clinicians do not control the conditions that produce these diseases in the first place. AI can integrate data about those conditions across a community and help identify the structural factors associated with poor metabolic health or low vaccine uptake, then generate evidence-informed options for residents, professionals, and policymakers to weigh. It can comb electronic records and claims to find the people overdue for screening or for the vaccines that produce lifelong benefit against cancer, disability, and even dementia, so that limited outreach goes where it matters most. The larger opportunity lies in moving from screening individuals to stratifying populations, directing community health workers and policy attention toward the neighborhoods at highest risk before disease develops.

Environmental and climate-driven threats

Climate change is increasing the frequency and severity of heat waves, wildfire smoke, and flooding, and expanding the range of the ticks and mosquitoes that carry disease. AI is well suited to anticipating these threats because it excels at forecasting. The same modeling that powers disease surveillance can integrate weather, air quality, and health data to predict where a heat event will do the most harm and to direct warnings, cooling resources, and air purifiers to the people most at risk. As the range of vector-borne disease shifts, AI can provide location-specific risk estimates, the kind of foresight that will matter as new tools arrive and the question becomes whether we use them, as with the Lyme vaccine now coming to market.

Two threads run through all four domains. The first is communication, where generative AI can take dense or disorganized material and turn it into clear, translated summaries tailored to a commissioner, a journalist, or a worried parent, provided a skilled human reviews every word. The second is the workforce: a program that today needs ten epidemiologists to clean and interpret data could use AI to move some of them into the field, doing the human work of engaging communities. That distinction, between extending people and replacing them, is where the genuine power of this technology will either be realized or squandered.