Episode 8 - 2009 - Robots - The Clever Things! (26.08.2021)
In 2009 the journal MaxPlanckResearch published an article about self-learning robots. These ingeniously designed machines learn to move without receiving any instructions from a control program. Similarly, robots, whose brains were developed by Nihat Ay and Ralf Der at our Max Planck Institute, are learning about their bodies and their environment. A scientific topic that is just as relevant today as it was then.
The science magazine MaxPlanckResearch gives an excellent overview of the activities of the Max Planck Society. It contains a wide variety of informative and easily digestible articles about research happening at the institutes. The magazine is published quarterly. Subscribe here to the print version of the MaxPlanckResearch magazine free of charge.
Prof. Nihat Ay headed the Information Theory of Cognitive Systems group as Max Planck Research Group Leader. In April 2021, he was appointed professor at the Hamburg University of Technology (TUHH) and Head of the Institute for Data Science Foundations as part of the Hamburg Innovation Port. He also holds a professorship at the Santa Fe Institute, New Mexico, USA, where he is involved in research on complexity and robustness theory. Since 2013, he is affiliated with the University of Leipzig as Honorary Professor.
Read more about Nihat Ay, his new position and his outstanding scientific career in our institute news.
Questions and Answers - Prof. Dr. Nihat Ay
Interview with Prof. Dr. Nihat Ay on the occasion of his inauguration as professor at the Hamburg University of Technology (TUHH) and head of the Institute for Data Science Foundations.
You will read a translation of an interview which was originally conducted in German by Franziska Trede from the TUHH Press Office.
You are now professor and head of the Institute for Data Science Foundations. How would you briefly and in an understandable way describe your field of research?
We live in an age where data is generated constantly and everywhere. The field of Data Science is concerned with extracting insights from this flood of data that can be used as the basis for informed decisions, whether these are made by individuals or global socio-political decisions. What is special about Data Science is that the process of gaining knowledge should be automated. After all, no human can look at all the data at once, let alone recognize patterns. Nevertheless, the capabilities of humans represent the natural model for this process. Humans are able to organize the stream of raw data of their senses, assign meaning to them and thus understand the world around them. This process of understanding runs almost playfully and represents the basis of all action.
How are you planning to set up the new institute? What will be the focus of your research?
The institute will take a holistic approach, unifying research on central aspects of learning intelligent systems. In particular, concepts and methods of machine learning, deep neural networks and embodied intelligence will be integrated. Here, mathematical theory building will play a central role and will be supported and guided by experimental work in a planned robotics lab. This two-pillar concept, on which my work so far has been based, will also be reflected structurally in the institute by me, the head of the institute, on the one hand, and the chief engineer on the other; I am currently looking for the right person for this job. The goal is to establish a focus for the field of embodied intelligence. Other test and application areas will be identified through collaborative initiatives and projects and used to develop foundations for Data Science.
What are you currently researching at TU Hamburg? Can you describe a concrete research project to me as an example, preferably explained in simple terms?
Like many of my colleagues, I am working on several projects at the same time. Each of them, however, represents only a puzzle piece within the context of a vision. One of those puzzle pieces deals with the interplay between supervised and unsupervised learning. The best understood is supervised learning, for which very efficient methods have already been developed. Consider the prime example of a child learning to recognize cats in pictures from her mother. In the beginning, the mother helps the child by providing the right answer, i.e., telling her whether or not there is a cat in the particular picture. After a few examples, the child should then be able to recognize the cats herself without help. Of course, this is only an example. Replace the child with an artificial learning system and the mother with an experienced medical doctor, and it becomes a matter of recognizing certain diseases, not the cat, on the basis of medical imaging techniques. Now, the big problem is that we are typically dealing with high-dimensional data where no one knows how to classify them, neither the, from the child's perspective, omniscient mother nor the experienced medical doctor, to stay with the mentioned examples. Thus, it is not about passing on existing knowledge, such as from the mother to the child, but rather about discovering knowledge. This is a central concern of Data Science. Within the holistic and mathematical conception of the Institute for Data Science Foundations, we pursue natural and promising approaches for this purpose. Here, fundamental ideas come from the important field of embodied intelligence, which is closely related to cyber-physical systems, a research focus of the TUHH.
How and where can your research or research example be applied?
As mentioned above, fundamental ideas for my research come from the field of embodied intelligence. Corresponding learning systems have a body that interacts with the world and thereby unfolds behavior. One can think here, for example, of a two-, four-, or six-legged robot that learns to walk. Controlling such a physical system can be very difficult if one tries to plan it from the outside. Yet this is the usual approach. Within our approach, on the other hand, the system develops its own unique view of the world on the basis of sensory data, from within, so to speak, and uses this for its goal-directed control. It turns out that this can often greatly simplify the problem of control, leading to robust solutions with low energy consumption. However, we want to go a step further and explore more general systems. Here we are dealing with complex systems that are composed of many interacting or mutually influencing components that do not necessarily form a coherent body, as in the case of a robot. Think, for example, of the traffic system, be it car, airplane or ship traffic, of social systems or, quite concretely, of the university system of the TUHH. Not only the legged robot should learn to walk or run, but also traffic and social systems should, in a more abstract sense, run optimally. What is the best way to achieve this? Our goal is to provide an answer to this with the fundamentals of Data Science.
What do you want to achieve with this? What contribution do you want to make to progress? (In line with the motto of the TU Hamburg: Technology for the People)
As the name of my institute suggests, it is first and foremost about fundamentals; I want to understand the process of data-driven understanding. How do we manage to comprehend the world around us from meaningless data? Or do we just construct our own world that is good enough to navigate us through life? The technical realization of artificial systems represents an important tool for developing the theory, in the spirit of Richard Feynman, an eminent physicist and Nobel laureate, who paraphrased this idea with the words "What I cannot create, I do not understand." The robotics lab of the Institute for Data Science Foundations aims precisely at creating systems that are capable of gaining insights from meaningless data and thus ascribing system-centric meaning to them. In the future, technical implementations will expand beyond their initial role as an instrument for basic research and be used in a wide range of applications involving the management of highly complex systems.
On which topic did you do your doctorate?
I did my PhD in mathematics, on a topic that describes geometric properties of learning systems. This is not so much about the geometry of systems in three-dimensional space, for example, whether the system has two, four, or six legs. Rather, it is about a geometry that describes how far, in an abstract sense, the learning system is from the learning goal. The field that deals with this kind of geometry is called information geometry. On the one hand, my dissertation, entitled "Aspects of a Theory of Pragmatic Information Structuring," was a kind of ticket into this field, and on the other hand, it enriched it by providing a new research direction. Meanwhile, together with three colleagues, I have written a book on the subject, which is considered to be a standard reference , and I have been Editor-in-Chief of the journal "Information Geometry" for some months.
Why did you decide to pursue a career in science? Did you always want to pursue this career path?
It was not a conscious decision, as one makes after weighing all the options. To be honest, I only became aware of the possibilities within the respective career step when it came up, so that one thing led to another and I am now the head of an institute at the TUHH. On the other hand, looking back, I realize that I was involved with scientific issues early on in my life, but without being aware of the fact that there is a profession where you even get paid for it.
What excites you so much about your work?
I am a mathematician. It fills me with great pleasure to see how a highly complex network of logical connections can emerge from simple abstract structures and grow into a far-reaching theory. Sometimes these connections follow our intuition, and sometimes they surprise us; logic does not necessarily respect our intuition. However, mathematics unfolds its real supporting power in areas where our imagination is too weak and intuition leads us to believe false things. Who can imagine a 1000-dimensional space? For most people, including mathematicians, imagination stops with the third dimension, the dimension that corresponds to our everyday visual space. All this is fascinating in itself. But the real wonder reveals itself when we can reduce the complexity of natural phenomena and processes to a foundation in terms of simple abstract structures, that is, when theory describes nature. Then we feel, at least I feel, that we understand how nature works. My dream is to gain such an understanding of those mechanisms that underlie natural intelligence. I firmly believe that it is the result of simple basic rules and unfolds in a variety of ways.
What qualities should one have to be a researcher?
As a researcher, one should be able to question things and overcome limitations of thinking. These are often the result of a cultural habituation process and exist only in our heads. My study of mathematics was ultimately a kind of therapy in which I learned not to be afraid of thinking the possible, even if it doesn't make sense at first glance. At the time, my mathematics professor roused me with the statement that 1 plus 1 can certainly be equal to 0. This is by no means nonsense. Today I know that such arithmetic operations form the basis for linear codes, which play a fundamental role in information and coding theory and are used in communication systems.
What would you like your students to take away with them?
I would like to encourage my students to get involved, really get involved, in subjects that are considered to be fundamental by experienced lecturers. It requires trust because we live in a world where knowledge is seemingly available everywhere and at all times.
What do you like about Hamburg?
Actually, I am new to Hamburg and have not had enough time to explore this city. However, the time has been quite enough to realize how friendly the people here are.
If you like, feel free to share a little private information about yourself, such as: Where did you grow up? Marital status? Hobbies?
I grew up in Bochum and studied at the Ruhr University in Bochum. I am married and have three grown-up children, two girls and a boy. Concerning my hobbies, I was once fascinated by painting, a long time ago, and I painted and exhibited by myself until graduation from high school, by the way with strong relations to my current research. My dream is to return to painting one day. Perhaps I will be able to realize this after building up the Institute for Data Science Foundations.