Where does AI fit into the big picture of public health?

In the view of University of Illinois public health researcher Bruno Nunes, artificial intelligence shapes up as a potent tool to predict and prevent public health problems, such as chronic diseases. 

But before AI-powered models are deployed in public health settings, they must be trained on richer data sets so they don’t amplify inequalities that exist in our healthcare system and society.

“To reach this future, we need to have better data to develop these models,” said Nunes, associate professor at the Department of Health and Kinesiology at the College of Applied Health Sciences. “Machine learning is a data learning process. It’s not just about using the fanciest algorithm—the problem is if you don’t have good data, you won’t have a good model.” 

Trained as an epidemiologist, Nunes is focused on public health strategies: how to prevent chronic disease and promote positive health outcomes at the population level, in some cases before people visit the doctor’s office. 

Artificial intelligence is already widely deployed in healthcare settings to better diagnose patients, especially reading medical imagery like X-rays, MRIs and CT scans.

With machine learning’s superior ability to detect patterns using huge tranches of data, Nunes envisions a future where models can accurately predict the risks of developing chronic diseases and allow populations to intervene earlier than before.

Part of this, Nunes argues, is AI may help us untangle “multimorbidity.” Many healthcare patients show up to the doctor’s office with two or more diseases, such as hypertension combined with diabetes or high cholesterol, which complicates management and quality of life. 

“Our health system and services are tailored to one disease. But in most cases, especially when talking about populational aging, most people are presenting different diseases at the same time,” Nunes said. “And the worst part of that is when we aren’t able to manage this patient well because they have such complex conditions and interactions.” 

His recent research has tested machine learning models on their ability to predict real-world outcomes. One recent study showed that machine learning models can predict a population’s dental service usage with solid accuracy but show poorer results with certain demographic subgroups. 

Nunes collaborated on a study that used an AI model to predict dental service use for adults in Southern Brazil. The model used 47 different characteristics—sociodemographic data, behavioral traits and oral and general health markers—to predict whether participants went to the dentist in the past year from a cohort study in Pelotas. 

Though the machine learning model’s predictions were largely accurate, it performed significantly worse across the board for mixed-race individuals in the study compared to Black and white participants, making the model unsuitable for real-world implementation in its current form.  

“None of the models are perfect: they present an error rate, and we need to deal with it,” Nunes said. “But if this error rate is higher for a subgroup of the population, the subgroup may be under- or over-diagnosed.

“If the model is not so good for people who already present with historical inequalities in the health system, the model can amplify these inequalities instead of decrease them.”   

Nunes tries to teach his students to frame the right questions in his new class, Artificial Intelligence in Public Health, which debuted in fall 2026 in HK.

Through critical discussions, he hopes to get students to think more about how “AI can fit into the big picture of public health,” and construct their own models around the right questions. 

“In most cases we tend to develop models for disease-related consequences or for problems which we already have an effective public health strategy, for example—but what if we could create equitable models to predict the problems in advance or issues without scalable solutions?” Nunes said. 

“You can’t just press a button to develop a machine learning model. You must have prior knowledge of the topic, skills and abilities to interpret the model considering public health principles. How can it be useful to solve the disease burden at the population level?”

Editor’s note:

To reach Bruno Nunes, email nunesbp@illinois.edu 

“Dental services use prediction among adults in Southern Brazil: A gender and racial fairness-oriented machine learning approach” is available online.  

DOI: 10.1016/j.jdent.2025.105929The database is publicly available: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/BTLAAD

Related news