"AI helps us to grasp more and more complex facts"

Let's go back in time a little. You first studied physics at the Technical University of Munich and later completed a Ph.D. in theoretical biophysics. How did you get into machine learning? 

My doctoral supervisor was a theoretical biophysicist, but my research focused on the memory capacity of “Hopfield networks”. These are a special type of artificial neural network. If you study such models, then you are essentially already in the academic territory of computer science. It's no longer just pure physics because it's not about inanimate matter, but about information processing. This area was not yet fully established in computer science at that time, but it is clearly part of the subject. Later in my career, I moved to Bonn and continued to work in the field of neural networks as an associate professor of practical computer science.

Why did you come to ETH? 
I was an associate professor in Bonn and had no prospects of being promoted there. At the age of 43, I got the opportunity to become a full professor at ETH Zurich. ETH had, as it does now, an excellent reputation, although the University of Bonn in Germany was also outstanding in mathematics, which was the home of computer science at the time. My wife and I had already built a house in Bonn, but as our children were almost finished with school, it was an obvious choice at that time. I probably wouldn't have moved to a less important university, as every move requires considerable personal resources.

What have you worked on in your research career? 
Even before I came to ETH Zurich, I was working on the question of how clustering algorithms assign their data to different groups.
The way this allocation works differs from that of classification algorithms. In classification algorithms, data is usually annotated manually by a human, and the algorithms are then trained with these annotations. For example, you want to automatically classify images of dogs and cats into two groups and use a training dataset to specify that an image should either be classified in the “dog” group or the “cat” group.

With clustering algorithms, there are no such labels, so there is no predefined “dog” or “cat” class. Nevertheless, the algorithm is supposed to eventually assign a label to every object. I wanted to find out how the algorithms carry out the clustering when there is no quality measure that they can use as a guide. I then applied this theory to various biological and medical projects. The approach of the clustering algorithms reflects the situation of a doctor who is tasked with predicting the probability of survival of a patient based on an X-ray image and other sources of information.

How has your field of research at ETH Zurich changed over the last 20 years? 
I had not foreseen that my field of research would develop so dramatically in the last 15 years. These are incredibly exciting times. The current rise of artificial intelligence affects all disciplines of science and, for me as a trained physicist, it is comparable to the introduction of quantum mechanics in physics. As a student, I wanted to know enough in my profession to see such a revolution emerge in my field of research and perhaps even contribute to it myself. It is an absolute stroke of luck for me that I was able to witness this time and actively participate in it as a researcher.

When I joined the Department of Computer Science, hardly anyone was interested in machine learning. Some professors, even in other research areas, did use machine learning in their research, but never to the extent that it is used today. And no one in the Department of Computer Science had made machine learning their core area. Things are different nowadays. There is now an Institute for Machine Learning with 11 professors.

Do you view these developments in the field of artificial intelligence with concern or enthusiasm? 
I'm not worried about the scientific development itself. My concern, if any, is that society may not sufficiently understand or anticipate the consequences of these scientific advances. New technologies – including artificial intelligence – have the potential to do a lot of good, but they can also be misused. Artificial intelligence is a technology that improves human thinking by enormously expanding the limits of the human capacity to store facts and grasp complexity. This is because the human brain tends to ignore details and focus on the big picture, i.e. to abstract. Enabling society to learn how to use these systems in an ethically correct way is an important educational task. New procedures need to be developed to ensure transparency, responsibility and fairness in the use of these programs.

At ETH Zurich, you were both a researcher and a lecturer and took on some administrative roles. How do you look back on your time as Vice Rector? 
I was Vice Rector for Study Programmes at ETH for four years. The role required a lot of time and empathy. But there are also incredibly interesting issues that come your way and that involve enormous responsibility. You are confronted with questions that are at the interface between the preconceived set of rules and an empathetic, ethically correct assessment of individual cases. The decisions you make can result in significant restrictions on someone's life options. For example, you must decide whether a student should be expelled from their programme. This must be grounded on a very good reason rather than the randomness of any given processes. The role of Vice Rector was certainly a challenge, but I think I was able to contribute reasonable solutions. My academic background has certainly helped me in this administrative role. Essentially, being Vice Rector is about monitoring the processes that take place. This has a lot to do with computer science, where we develop automated processes that are then executed by the computer.

Is there anything you have learned during your time as Vice Rector? 
First and foremost, I became a scientist to do research. However, in addition to producing new knowledge, as a university lecturer I also have the responsibility to pass on existing knowledge. During my time as Vice Rector, I learned that the university's priority is always teaching, and that research comes second. However, as the quality of research is easier to measure, it is often given more importance than teaching. Students at universities should first and foremost be trained to become intelligent problem solvers who can make reasonable decisions even in conditions of great uncertainty – regardless of whether they go on to work in industry or stay in academia.
Something else I have learned in this context: everything we do is a service to society. This is true of administrative tasks just as it is of research and teaching. In this sense, I believe that all three areas should be valued equally.

You are retiring in July. What will you miss most in retirement? 
When you've been in a place for a while, you tend to long for that familiar environment and the community of colleagues, doctoral students, postdocs and students. I will certainly miss many things, but I don't think I will suffer from it. This is currently an optimistic assessment and if you ask me again in a year's time, my opinion may have changed. But I am convinced that new opportunities will arise in the future.

Do you have any concrete plans? 
My family is relatively large. We are expecting our eighth grandchild soon. I'm sure I'll have a few tasks ahead of me. Professionally, I haven't prepared myself for a direct follow-up job and I'm not actively looking for one. However, I would like to maintain my contacts with the institute and try to make myself useful as an emeritus professor. I also think that I will continue to do research, but probably less than now. I would also like to contribute my time and expertise to public relations work to support society in this digital transition.