AI in Medicine: Much More Than Just a Language Model
October 6, 2026
Artificial intelligence can analyze medical images and signals, detect abnormalities, process large amounts of data, and integrate different types of information. But its potential goes far beyond that. Prof. Dr. Andreas Maier, director of the Pattern Recognition Lab at Friedrich-Alexander University Erlangen-Nuremberg (FAU), conducts research on machine learning methods and their application in medicine. In this interview, he explains why medical AI should take a holistic view of people, why data alone is not enough—and how AI could help relieve the burden on doctors and free up more time for them to provide personalized care to patients.
Professor Maier, what are you and your team working on at the Pattern Recognition Lab at FAU?
Prof. Dr. Andreas Maier: We study patterns and their recognition. Essentially, our work focuses on developing machine learning methods. Today, this would likely be referred to as artificial intelligence. However, our lab was founded back in 1975—long before machine learning became the major topic it is today.
The terminology has also evolved over time. While pattern recognition originally stemmed more from electrical engineering and signal processing, the term “machine learning” has become established in computer science. In the field of databases, on the other hand, the term “data mining” was used. These research areas initially developed independently of one another, but today they essentially use the same methods and have increasingly merged.
Our lab has grown significantly over the past 50 years. Today, we work in eight research groups on both fundamental methods and various fields of application. The spectrum ranges from natural language processing, emotion recognition, and computer vision to medical systems and medical AI, which represent a key focus for us.
Many people associate artificial intelligence primarily with chatbots like ChatGPT, which are based on large language models. In medicine, however, the potential of AI extends far beyond language processing. Why is it important to think more broadly about medical AI?
Prof. Dr. Andreas Maier: In medical AI, we are ultimately trying to understand the patient. A CT scan, an EEG, or an ECG always depicts only a specific aspect of the patient and their state of health. An ECG, for example, provides information about heart activity, while an EEG provides information about brain activity. However, a single source of information is not enough to truly understand a disease. Even doctors don’t just look at an image or a measurement. They talk to patients and incorporate their symptoms and perceptions into the diagnosis.
We should take the same approach with AI. It’s not enough to look at just one sensor, one image, or one piece of text. We need as holistic an understanding as possible. This will enable us to develop better methods—from diagnosis to treatment to follow-up care.
Important clues in medical data are not always immediately apparent to humans. How can AI help identify even very subtle patterns and changes?
Prof. Dr. Andreas Maier: AI and machine learning are particularly good at recognizing recurring signals, structures, or patterns. This can be helpful, for example, when a disease is in a very early stage and changes are initially only faintly detectable.
At the same time, we must be cautious. An algorithm can identify statistical correlations without understanding what actually lies behind them. Here’s a simple example: Shoe size allows us to infer height. Since men are, on average, taller than women, shoe size can statistically provide clues about gender. The actual relationship, however, is more complex. This is precisely where the challenge lies: AI initially identifies correlations but does not automatically understand the underlying causes. That is why close collaboration with doctors and medical experts is so important. They can assess whether a detected pattern is actually medically relevant.
One focus of your research is medical imaging. What potential do you see for AI in this area?
Prof. Dr. Andreas Maier: There is great potential in reducing the workload associated with routine tasks. AI can perform certain analyses very quickly, thereby giving doctors more time for their patients.
In addition, AI can process very large amounts of data. This could be helpful, for example, in the case of rare diseases. A doctor may have encountered a particular rare disease only once in their daily clinical practice or may know it only from textbooks. AI systems could learn from data collected from different institutions and regions and make this knowledge available. At the same time, the training data must reflect human diversity as accurately as possible. For example, if algorithms are trained exclusively with data from a specific region, they may work better for the population in that region than for other groups. International collaboration can therefore help develop more representative systems.
AI systems learn primarily from large amounts of data. Should medical AI also incorporate expert knowledge—and what difference might that make?
Prof. Dr. Andreas Maier: Large amounts of data are important because we can train AI models using many examples, which generally improves their performance. Experts, however, take a different approach: They try to understand the underlying relationships and describe them in models that are as concise as possible. It is precisely this connection that AI still lacks to some extent today.
An important next step will therefore be to link data-based models more closely with expert knowledge. In medicine, for example, existing expertise could be used to verify AI results and ensure that they are consistent with medical findings. Looking ahead, AI should not only make predictions but also be able to identify underlying relationships and mechanisms and describe them in a way that humans can understand. I see this as an important area of research for the coming years.
In your view, what is often overlooked in the public discussion about AI in medicine?
Prof. Dr. Andreas Maier: When we talk about the use of AI in medicine, the focus is often on safety, robustness, and reliability. Of course, these are important. However, we shouldn’t focus solely on potential risks, but also on the opportunities and concrete benefits for patients.
We often focus primarily on the risk of introducing a new method. But we should also ask what the risk is of failing to further develop a helpful technology. For a balanced assessment, we simply have to consider both sides, because AI-based methods can, for example, support diagnoses, improve treatments, or reduce the workload on doctors.
If AI in medicine develops in the right direction, what specific changes should its use bring about in the daily lives of patients and medical staff?
Prof. Dr. Andreas Maier: Our healthcare systems are under enormous pressure. AI won’t suddenly make medicine cost-effective. However, it can help automate certain tasks reliably, thereby freeing up time.
Analyses, simulations, and other processes necessary for a thorough diagnosis could increasingly take place in the background. My hope is that this will free up more time for personal conversations, better care, and human interaction.
Thank you very much for the interview, Professor Maier!
