Dr. Nils Schweingruber is a co-founder of IDM. The UKE subsidiary is a non-profit limited company, distributes its own language model for writing doctors' letters and has developed its own speech recognition system. In this interview, the neurologist talks about the importance of technological sovereignty, Hamburg as an AI location and the possibilities of artificial intelligence in medicine.
ARIC: Who are you and what do you do?
Dr. Nils Schweingruber: My name is Nils Schweingruber and I am the Managing Director of IDM gGmbH, a wholly owned subsidiary of the University Medical Center Hamburg-Eppendorf (UKE). We develop AI applications for German-speaking countries and are a non-profit limited company. We want as many people as possible to be able to use this technology in their everyday clinical work.
Why did you decide to work as a non-profit limited company?
We are convinced that AI applications based on health data collected in a spirit of solidarity should also benefit the healthcare system. By licensing our models to cover costs, we are financing further developments in this area. In Hamburg, for example, we have created a place where AI development actually takes place. We have set up a team of software developers, AI specialists and experts in medical data and regulation. We also have the right infrastructure: we can train AI models locally directly on the clinical data and bring them directly into clinical operation and testing. This is a major advantage because we can test a wide variety of applications directly in everyday clinical practice. As soon as we see that an approach delivers real added value, we bring it to our other customers.
What specific models have you developed?
We currently have two products. One is Argo, an LLM (Large Language Model) that has been trained on the entire text of the UKE and is used for writing doctors’ letters. On the other hand, we have developed a speech recognition system: Orpheus. It demonstrates very well the added value of advancing this technology here in Germany. We can either operate Orpheus locally on the servers of large institutions or offer it via our sovereign cloud, making AI technology available across all sectors in the entire healthcare system. Given the enormous documentation burden, reliable speech recognition is a real benefit. We have been on the market with Orpheus for almost a year and are now in use in four university clinics, over thirty hospitals and more than 200 surgeries. For us, this is a strong signal of how important sovereign AI development is. The solution is so well received because it has been trained to be highly specific to German medical terminology and now outperforms many standard products on the market. This specialization is directly reflected in a significantly higher quality.
What are the specific things you have to consider in the German market?
Our take is: Generalist open source models or market-leading language models are well suited for everyday issues. However, they are not sufficient in highly regulated and very specialized domains such as medicine. Here, specialized models that have been trained on the relevant processes and data are needed to deliver the expected quality.
Further developing this technology on site and continuously improving and fine-tuning the models with new data creates particular added value. It is precisely this process that we fully map here in Hamburg, without having to rely on external APIs.
What data is the model trained on exactly?
Initially, the model was trained with data from the UKE. However, we now also use data from other university clinics, hospitals and the outpatient sector. This gives us a very broad data set, probably the most comprehensive speech-to-text data set for German medical language.
The “AI and Medicine” specialist group at ARIC works on issues relating to artificial intelligence and the use of medical data and meets regularly for roundtables.
Medical professionals who want to think outside the box, help shape digital technologies and actively contribute to the transformation of medicine are invited to attend.
You can find out morein the corresponding call.
You are a doctor. When and how did you become interested in AI?
I am a neurologist and have worked in acute medicine and intensive care for a long time. Enormous amounts of data are generated there, as critically ill patients are monitored around the clock with a wide variety of sensors. Some of them are in a deep artificial coma and are ventilated.
I asked myself at the time, What actually happens with all this data that is generated here? We should actually be able to use this data to develop forecasting models, similar to a medical weather forecast. Back then, I started to make healthcare data usable for prognosis models and quickly realized that the potential goes far beyond that: data can help to simplify, streamline and partially automate clinical processes. This was one of the origins of the IDM, which emerged from a group of committed doctors. We were and are convinced that this technology must be developed in Germany itself. Otherwise, we run the risk of it being imposed on us by large corporations at some point. At the UKE, we have a special starting point for this, as we have been a fully digitized hospital since 2009. This gives us a very good basis for developing AI models and then transferring them to other areas of the German healthcare system.
How has the IDM developed since it was founded?
There are now 18 of us and we also work with over ten medical students. This year, we plan to grow further, particularly in the field of speech recognition, but also in the AI-supported structuring and utilization of data. We are also planning to bring our Argo language model to more clinics. Another important step is certification for the manufacture of medical devices. This is a very extensive process that we are tackling this year in order to be able to offer AI models for diagnostics and care in the future.
What kind of products might you be launching on the market?
You can imagine an incredible number of things. It is conceivable, for example, that LLMs or AI agents will support doctors in everyday clinical decision-making processes in the future.
“With the IDM, we want to make a contribution to the sovereign development of AI models in Germany and Europe and also make them available to other start-ups.”
What would you like to see from clinics and perhaps also doctors when dealing with AI?
I always emphasize the importance of a conscious approach to AI products. There are many exciting start-ups with great ideas, but the majority of them only use market-leading language models and are ultimately nothing more than wrappers.
These start-ups are dependent on the resource of AI models, which they mainly obtain from outside Europe. With the IDM, we want to make a contribution to the sovereign development of AI models in Germany and Europe and to make them available to other start-ups. I would therefore like AI users to take a closer look at what data they enter into which models. From our perspective, sensitive patient data does not belong in generally available language models.
What does technology sovereignty mean to you?
For us, technological sovereignty means actively shaping technology. This requires investment in know-how, qualified people and efficient infrastructure. This is exactly what we at IDM work on every day.
At an infrastructural level, sovereignty also means that we in Germany and Europe need our own powerful models in order to create real alternatives to the large AI providers from the USA and China.
Especially in the current political situation, it is very important to take a close look. AI has enormous potential, but if we only consume it, we are not only paying with money, but also with data. This data provides valuable knowledge about our society. This must not be relinquished lightly, especially when it comes to health data.
I am convinced that we are currently well on the way to building a sustainable infrastructure, for example by expanding large data centers. It is now crucial to fill this infrastructure with our own European models and not leave it exclusively to international technology companies. We must ensure that the data processed there is not misused for other purposes.
Health data is some of the most sensitive data of all. At the same time, we are already seeing patients making extensive use of market-leading voice models, for example to obtain a second opinion or advice. However, in addition to positive individual cases, there are also numerous reports of incorrect or potentially dangerous information. From a medical perspective, caution is therefore advised here. We should be careful that our society does not become completely transparent and that we do not become any more dependent on technology companies. We must now fill the existing infrastructure with our own European models and ensure that sensitive data is not misused.
What does the Hamburg location mean to you in this context?
We are very interested in strengthening the region and advancing AI in Hamburg in a targeted manner. I would like to take this opportunity to promote this. The conditions are very good and the potential is enormous. What is needed is a joint effort to exploit synergies and further develop Hamburg as an AI hotspot.
“We have a lot of talent here”
Who knows, perhaps political developments are now moving so fast that we are forced to do so.
Unfortunately, it is a recurring dilemma that we often lag behind in our system. In principle, it is also okay that democracies are sometimes a little slower. However, when it comes to technological leaps, this becomes a problem because we don’t adapt quickly enough. It’s often the basics: we still don’t have fiber optic connections everywhere here in Hamburg. You need a stable internet connection and a powerful infrastructure to be able to operate and train large AI models. These are issues that have been known for decades. Nevertheless, I am hopeful and optimistic that we will see good developments this year and next.
Where do you get your optimism from?
I believe that the wind is currently in our favor. In Europe, we need to re-establish the idea that we need to stick together and not hide. We have a lot of talent here and a lot of technology that is being created here. We just need to make consistent use of it for our society. The potential is huge.
At ARIC, we meet a lot of people who are doing it right. That motivates us and also at work.
Sometimes we have to dare a little more, just take a chance. Then it will all work out.
Interview: Sabrina Pohlmann
With our interviews, we want to introduce you to different perspectives and players in the field of AI. The positions of our interview partners do not necessarily reflect the positions of the ARIC.
Further interviews:
