“AI will not replace humans – but will surpass those who don’t use it”

Artificial intelligence is on its way to becoming an integrated part of health research, but how is the technology used in practice, and what challenges come with it?

We spoke with Adam Hulman, Associate Professor at the Department of Public Health and Steno Diabetes Center Aarhus, about his work with machine learning, prediction models, and the ethical considerations associated with AI in healthcare.

How did you become interested in artificial intelligence?

I have a background in mathematics and have worked with classical epidemiology for many years. My interest in AI was sparked about four or five years ago when I watched a documentary about how DeepMind developed an AI that could defeat the world’s best Go players. I was fascinated because Go is an extremely complex game with more possible moves than atoms in the universe. If AI could master Go, I thought, it must also have potential in health research. So, I began exploring how machine learning could be used to improve prediction models in diabetes research.

How do you use AI in your research?

I lead a group at Steno Diabetes Center Aarhus, where we work with machine learning and algorithms to develop clinical prediction models. We focus particularly on cardiovascular disease in people with type 1 diabetes. We aim to integrate different types of data—such as eye scans, glucose sensor data, and voice analysis—to identify patterns that may be linked to diabetes complications.

What can AI contribute that traditional methods cannot?

I prefer to call it machine learning, but yes, we work with AI because classical statistical methods are not designed to handle the large and complex datasets we have today. For example, we can now incorporate real-time data from glucose sensors, which measure blood sugar levels every five or fifteen minutes. This opens up entirely new possibilities for prediction, but we need machine learning to make sense of such data.

How far are we from using AI in clinical practice?

AI is already being used in some areas, such as identifying healthy structures in scans, like bones and organs. But when it comes to prediction models for diseases like cardiovascular disease, we are not quite there yet. We have developed a model and are now working on expanding it with more data types. However, many prediction models exist in the literature, and only a few are actually implemented. We are trying to ensure that our models not only work in a research context but also make sense in clinical practice.

How can AI support doctors?

Doctors will be able to use AI to assess a patient’s risk of, for example, cardiovascular disease. The more data we can incorporate, the better the models we can develop. This provides a consistent and fast starting point that doctors can then fine-tune. It saves time, allowing resources to be allocated to other tasks, such as patients who need to start treatment. However, AI should not replace doctors—it should assist them. Many people believe AI will replace humans, but what will actually happen is that AI will surpass humans who do not use AI.

How do patients perceive the use of AI in healthcare?

We have studied this, and it depends on the context. If AI is used for advice on lifestyle or diet, people are very open to it. But when it comes to acute situations or serious diseases, people become more sceptical. Who is behind the technology also matters. If the technology is developed by a public institution like Steno or Aarhus University Hospital, there is greater trust compared to a random app from the App Store.

How can Denmark position itself in the AI race?

Denmark has a huge advantage with our health registries, but we need to find a balance between data security and innovation. If we become too rigid with regulations, we risk falling behind countries like the USA and China.

What is the next step in your research?

We will continue integrating new data types into our prediction models and exploring how we can implement AI in a way that makes sense in clinical settings. We are also working to better understand the patient perspective—because if patients do not trust the technology, it will never be used.