Complex Computer-Designed Quantum Experiments

Quantum effects are very difficult to understand intuitively. An example is the so-called quantum entanglement. There, it seems that two particles remain connected over large distances. When you measure the first particle, something happens to the second particle instantaneously. Effects like these are the basis for a new class of technologies, with a variety of applications. For example, they would allow to ensure completely secure communication; to implement novel and much faster models of computers or to improve imaging systems such as telescopes or microscopes. Many of these quantum effects are also an important part of basic research. To understand these phenomena in more detail, one would like to be able to create and examine them in laboratories. However, that is often a problem: designing the right experiments and setups can prove to be a very difficult task. Only recently have automatic search algorithms been used for such questions, and some of these computer-suggested experiments have already been successfully implemented and investigated in laboratories. However, as there are a tremendous number of possibilities for experimental setups, this method is limited. Interestingly, there is a field of research that, although at first glance far removed from quantum optics, has very similar questions. In the field of quantum chemistry one wants to design novel molecules and materials that have particularly desirable properties, such as for new batteries, for efficient photovoltaics, and many others. For some years now, state-of-the-art artificial intelligence algorithms have been used to help design new materials. In my project, in the group of quantum chemist Alan Aspuru-Guzik, I will translate these A.I. algorithms from chemistry to quantum optics. This gives me the opportunity to answer many of the open questions of quantum optics. Among other things, these questions deal with the generation of very high-quality quantum states, which are crucial for quantum computers. Furthermore, I will use my algorithms to design concrete quantum-based enhancements of optical images in astronomical telescopes. These computer suggestions can then be studied in laboratories and can lead to considerable scientific and technical advances. Finally, I will explore what the underlying reasons and principles are, that these new solutions work, which until now have been hidden from human researchers. Findings from this could have far- reaching implications for the understanding of quantum optics, as well as for quantum chemistry, and provide new understanding or mathematical descriptions that were unknown until now.

Human intuition, which guides us perfectly through our everyday world, fails completely when dealing with the world of quantum physics. The ideas of superposition (two properties of a system that appear to be realized simultaneously) or the phenomenon of entanglement (two systems, such as photons - i.e. individual light particles that appear to be connected over large distances) are just two examples of effects that we never experience in our everyday life. Now, of course, the question arises whether this human intuition is actually the best way to achieve scientific and technical progress in quantum physics. For example, can human scientists actually design the best and most interesting quantum experiments, or are there other methods that work better? In my research project, I dealt with how we can use artificial intelligence to create new experiments in quantum physics - and there especially with light particles. My approach was this: I went to Toronto to join a computational chemistry research group. In chemistry and materials science in general, and in my host group in Toronto in particular, remarkable artificial intelligence algorithms have been developed for the design of new functional materials, such as new batteries or photovoltaics. In general, the incorporation of AI methods in chemistry was (and is) far advanced and much more developed than in physics. So my plan was this: learn as much as possible about the methodologies of AI in chemistry, and transfer it to physics and quantum physics. In particular, I am interested in the AI-based development of new quantum experiments and quantum hardware. Back to quantum physics. During that time, it was actually possible to develop an algorithm that designs realizable quantum experiments. It is possible to do this many orders of magnitude faster than was previously possible. In addition, we managed to implement the algorithm in such a way that we can learn new ideas and concepts from the results. It has allowed us to answer some open questions in experimental quantum physics - like how to develop certain resources for quantum computers. In addition to the pure solutions, we also understood the underlying principles - and thus one of the first examples in which a person learns new ideas from artificial intelligence in the natural sciences. Finally, I would like to point out a work in which we are fundamentally concerned with how people can gain new understanding through AI. In doing so, we accessed interviews from more than 50 physicists, chemists and biology researchers, as well as new ideas from the philosophy of science and AI research. In the future, this research could radically change how human and artificial scientists work productively together to answer questions about our universe.