By Dr. Marco Vinicio Benavides Sánchez.

Imagine your country has a limited health budget. How do you decide what to invest in first? Hospitals, vaccines, mental health care, disease prevention? And most importantly: how can you tell if those decisions truly improve population health?

For years, this has been a complex, slow, and costly task. But now, a group of researchers proposes an innovative solution: using artificial intelligence to better understand how public health funds are distributed—and doing so with greater precision, speed, and transparency.

🧠 What Are Scientists Proposing?

Daniele Guariso, Rilwan Adewoyin, Gisela Robles Aguilar, Omar A. Guerrero, and Alisha Davies have developed a system that uses large language models—like those powering tools such as ChatGPT—to help plan public health spending. Their proposal was recently published in the journal Artificial Intelligence in Medicine.

What makes this approach special is that it not only automates complex tasks, such as linking budget items to health indicators, but also incorporates something fundamental: a measure of uncertainty. In other words, it doesn’t just say “this might work,” but also “this is more or less reliable.”

🏥 Why Is Health Budget Planning So Difficult?

Public health doesn’t depend solely on hospitals or medications. It’s also influenced by factors like education, transportation, housing, and the environment. That’s why experts insist on a holistic view: improving health means understanding how different sectors interact.

But putting that vision into practice is challenging. Governments must decide how to allocate budgets and link each expense to concrete health outcomes. This process, known as budget tagging, is usually done manually by policy and health economics experts. The problem? It’s slow, expensive, and can be biased by personal opinions.

🤖 CPUQ: A Tool That Thinks… and Doubts

This is where artificial intelligence comes into play. The proposed system, called CPUQ (Categorical Perplexity-based Uncertainty Quantification), uses language models trained on millions of texts to analyze budget descriptions and relate them to health indicators.

But CPUQ doesn’t just make predictions. It also estimates how confident those predictions are. Why does this matter? Because in public health, making decisions based on unreliable data can have serious consequences. Knowing when a model “doubts” is just as important as knowing what it “believes.”

📊 How Does CPUQ Work?

Imagine a map where each point represents a budget item (like “primary care” or “hospital infrastructure”) and another point represents a health indicator (like “infant mortality” or “access to medical services”). CPUQ creates connections between these points using natural language descriptions.

What’s fascinating is that these connections aren’t arbitrary. When compared to those made by human experts, the results are surprisingly similar. And in some cases, CPUQ identifies subtler relationships—like how child poverty can affect access to medical care—something other AI systems struggle to detect with such precision.

🔍 What Does “Uncertainty” Mean in This Context?

Most current AI models tend to be “overconfident” in their answers, even when they’re wrong. CPUQ changes that logic. Instead of giving a single answer, it generates probabilistic distributions—a way of saying “this is what we believe, but here’s the margin of error.”

This transparency is key. It allows policymakers to make better-informed decisions, understand risks, and place greater trust in AI for sensitive areas like health.

🌍 Where Could It Be Applied?

CPUQ’s potential is vast. Governments, international agencies, and NGOs could use it to improve resource allocation, especially in countries where time and money are limited.

Moreover, since the system is based on natural language, it can easily adapt to different languages and institutional contexts. This makes it an ideal tool for advancing the global goal of “Health for All,” especially in resource-limited regions.

💡 Why Does It Matter?

In a world where health systems face increasingly urgent crises—from pandemics to climate change and social inequality—planning isn’t enough: we must plan intelligently, swiftly, and transparently. CPUQ doesn’t just speed up analysis—it sheds light on it.
And most revealing of all: it shows that artificial intelligence doesn’t have to be a black box that imposes answers. It can be an ethical compass, a critical tool that helps us not just decide—but decide better. More than technology, CPUQ is an invitation to think humanely in times of algorithms.

---

To read more:

• Guariso, D., Adewoyin, R., Robles Aguilar, G., Guerrero, O. A., & Davies, A. (2025). A generalized LLMs framework to support public health financing through probabilistic predictions and uncertainty quantification. Artificial Intelligence in Medicine, 168, 103203. https://doi.org/10.1016/j.artmed.2025.103203

#ArtificialIntelligence #Medicine #Surgery #Medmultilingua