TUM: Prof. Schoellig, Prof. Rückert, let's begin with a general question on Artificial Intelligence (AI): What can it do better than people can?
Daniel Rückert: More than anything, AI can process much larger amounts of data than humans can. This is enormously beneficial in my specialty area, the use of AI in medicine. Thus for example AI is capable of sifting through all radiological examinations performed worldwide to identify patterns which make it possible to find very rare illnesses based on current x-ray images of patients which might otherwise go unnoticed, even when examined by highly skilled physicians who may not have had the chance to gain any experience with the illness in question.
TUM: That's a lot of responsibility for Artificial Intelligence. It can be a matter of life and death...
Rückert: True, but the life and death decision is never made by AI; AI simply serves as a tool for the physicians and recommends promising treatment methods. In many areas AI can achieve a very high rate of accuracy in diagnosis, in some cases higher than that of human medical experts. But of course the AI also has to be trustworthy and appropriately tested. If you ask the patients, they always say: I want to know what the AI found out!
TUM: Prof. Schoellig, your specialty area, Machine Learning and Autonomous Driving, is also concerned with safety.
Angela Schoellig: Yes, but with vehicles and robots that move in our environment, an additional factor is that the respective environment constantly changes and the machines have to be able to react accordingly. So we're not only talking about recognizing patterns in enormous volumes of data, but also about rapid and adaptive actions. In the past robots were only able to do exactly what they had been programmed to do. Machine Learning and the algorithms we're developing will make it possible in the future to complete highly complex tasks, for example safely steering a car through chaotic inner-city traffic – regardless of what the weather outside is like.
TUM: How do you think this will be possible without any errors?
Schoellig: Guaranteeing safety is extremely difficult; human actions are not always error-free either. One important aspect of my research is integrating safety questions as a permanent component of the software and hardware and then making them actually work out in the real world.
TUM: A question for you both: What is your greatest research objective? Where is it all leading?
Rückert: I have two major goals: General practitioners are the most important persons of trust for people with health problems, but as physicians they can't possibly have perfect knowledge of every possible complicated and specialized topic. AI should help them make the right diagnosis as early as the first contact with the patient and help them initiate the right treatment. And my second major research goal is to use AI to make personalized treatments and medications available to every patient. Every person is different and reacts individually to different active ingredients. Genetics plays a major role here as well, a highly complex aspect associated with enormous amounts of data. This is a great playing field for AI to demonstrate its strengths.
Schoellig: I'm especially interested in the implementation of Machine Learning and Robotics in products for everyday life. We're talking about the development of safe and efficient service robots that can be used for example in storage facilities or in hospitals and supermarkets. They can autonomously fill shelves, support care staff in serving food or can put clothing back on the hanger after it has been tried on in clothing stores. Robotics will make many people's lives much easier.
Prof. Angela Schoellig most recently conducted research at the Dynamic Systems Lab of the University of Toronto. In addition to theoretical work, she also addresses practical applications like self-driving cars. She earned her doctorate in 2013 at ETH Zurich and has already won numerous awards and subsidies, recently for example the Canada CIFAR AI Chair of the Canadian Institute for Advanced Research. Schoellig is now TUM Professor for Safety, Performance and Reliability of Learning Systems.Her chair is also an endowed chair that is financially supported by Infineon Technologies AG.
Prof. Daniel Rückert studied information technologies at Technische Univeristät Berlin. He received his doctorate in 1997 at Imperial College London where, before coming to TUM, he was Professor of Visual Information Processing and Dean of the Department of Computing. Rückert has developed ground-breaking methods for using Artificial Intelligence to generate particularly meaningful images from computed tomography images and magnetic resonance images, analyzing them and interpreting them for improved medical diagnostics.