Interview with Henrik Junklewitz. He received a diploma in physics in 2009 and a Ph.D in physics in 2014, both from the Ludwig-Maximilians-Universität München. He has been a post-doctoral researcher in statistical inference, machine learning and imaging for astrophysics at the Friedrich-Wilhelms-Universität Bonn. Currently, he is a scientific project officer for machine learning with the Joint Research Centre of the European Commission. His current research interests focus on robustness, security and explainability of machine learning, Bayesian statistics and inference, and applications of machine learning in cybersecurity.
Do you consider your work to be art or science? If you perceive some of your work as scientific research and some as art research, how much of what you do is science and how much of it is art?
Henrik Junklewitz: I would not consider much of my daily work as a researcher to be art. Of course, that depends on the definition of art. There is a formal academic definition (which I am not very familiar with), but it really comes down to what art means to you. It is a very subjective notion. For me personally, art is something I associate more with creative craftsmanship, which can be, but does not necessarily have to be, based on intellectual discourse. Thus, in that sense, I would consider some of the more explorative elements of research to be related to art, for instance searching for and pursuing new ways to solve problems, experimenting with computer algorithms and even the elements of storytelling involved in good science communication and in writing papers. The biggest difference for me is that I do not believe that art has to abide by the same intellectual and sometimes rigid rules as research; art can just be created for its own sake, its beauty, its creativity, or for no reason at all – maybe you just want to pursue a certain craft. I am an amateur musician too, and often creating music is exactly like that for me. I would never carry out all parts of a research project in this manner.
Should we get excited if a computer that learnt patterns from a training set can generate new artefacts with the same patterns? Or should we aspire to have a computer make its own patterns?
Henrik Junklewitz: I am not convinced that I fully agree with the underlying premise of this question. I would argue that it is hard to define the notion of ‘a computer’s own patterns’, since most patterns I can think of are made up of more basic patterns that, of course, are ultimately a basic part of our natural environment, or at least of our ideal mathematical representation of it. Ultimately, what the capability of modern AI systems to reproduce elaborate patterns means to me is that we, as humans, are in the process of learning that, with enough repetition, many more tasks can simply be learnt and reliably carried out than we previously thought. I find that intriguing in itself. A human artist also learns their craft through repetition, so maybe what we perceive as a new pattern is only an intuitive composite of these learnt things. In that sense, I would be confident that an AI system could eventually, with enough training, combine the patterns it reproduces into something that we as humans perceive as new.
Throughout history, works of art have shared similarities tied to the time and place in which they were crafted. Ornaments, materials, musical instruments and pigments were shared across entire continents. In various ways, our culture is the product of what was passed on to us by others who modified it a bit and added something new – a sort of mash-up. AI art basically goes through this human filter: humans are responsible for developing the networks, the algorithms and the machine learning. When will AI art be able to stand on its own, without the human element?
Henrik Junklewitz: I think this is unlikely to happen anytime soon, if ever. We are far from reaching the goal of real general artificial intelligence. At this point in time, AI is mostly a large collection of tools designed by humans for specific purposes. It’s true that we are living in a time where these tools are becoming extremely powerful and successful, even outperforming humans in specific tasks, such as playing a game or detecting a very specific pattern in data. But it’s also important to realise that the term ‘artificial intelligence’ can be a bit of a misnomer, largely because of the many meanings people associate with the word ‘intelligence’. To put it in layman’s terms, contemporary AI is capable of performing tasks that require limited cognitive reasoning: it is given clear information about the problem, rules for decision-making and ways to process new data, and it never steps outside of these boundaries. All of these tasks are meticulously designed, curated and prepared by humans. They decide what information will be encoded in the AI, program the rules and determine how the data is processed. I think, in general, it is much more useful and also less dangerous to consider AI as a powerful new tool for humans rather than something that is self-aware or capable of agency. Less dangerous, because it makes us aware that potential problems with AI are ultimately due to human factors, such as biases in automated decision-making. Useful, because this is an empowering thought for people who are being told that AI will change their profession. This is what happens in art: we embrace new technologies and make use of them to produce new products, get different results and create new art pieces. We have done this with new technologies countless times in the past.
Do you feel that AI artists are merely copy painters that mimic techniques and methods of artistic expression in a pre-determined artistic medium?
Henrik Junklewitz: Yes and no. Following on from my previous answer, I would not consider an AI system to be an artist at the moment. And AI tools used to produce art on their own will of course simply be reproducing elements that they have been trained to produce or identify. So to me, AI is exactly the same as any other tool used by an artist, like a paintbrush. If an artist using an AI tool does something interesting, creative, captivating or novel, I don’t mind the fact that the tools themselves are just reproducing patterns. I think what is likely to happen is that AI will come to produce mass-produced, cheap designs, that can be used for all kinds of purposes, but be considered to be on a completely different level to human art, much like a chair mass-produced in a factory according to an industrial design is very different to a chair that was handcrafted by an interior designer.
Do AI and virtual or augmented reality have the power to alter the structures of the contemporary art world?
Henrik Junklewitz: I think so, in the sense that in many other domains AI can be a truly transformative technology, when considered as a tool to be used by humans. I don’t think that the art world can or would want to evade those transformations. But I consider the world of art to be a very open-minded place and I am very confident that artists will happily embrace these new technologies.
Can an AI artist step away from the human forms of artistic expression (including virtual reality (VR) art, video and robotics) and create its own forms of artistic expression?
Henrik Junklewitz: Again, my position would be that this is unlikely to happen anytime soon. Of course, an AI system could produce works of art that are different to previous forms of human artistic expression. But I am not convinced that this will be any different to past technologies that entered the artistic world and opened it to new forms of expression, like the electric guitar and video art.
How will science influence art in the age of the internet and VR, and in what fields predominantly?
Henrik Junklewitz: I think for the moment, the biggest impact – as we have already discussed in this interview – is in fact from developments in computer science and AI, especially given that we are in the age of the internet and VR, as you mentioned. At first, that influence will impact digital art, of course, or all art that can now also be created digitally, for example producing realistic and artistically crafted pictures and images, or composing and recording music digitally. But with the advent of cyber-physical systems, the internet of things and new developments in robotics, these developments in computer science and AI are likely to start leaking into physical art as well, and – as it has already started to happen with VR – I expect the boundaries to get increasingly blurred. I also think that modern biotechnology has the potential to influence human art in the 21st century, but I am no expert in this matter.
Can AI outperform an artist in traditional forms of artistic expression, i.e. manual art forms such as painting and sculpture?
Henrik Junklewitz: Again, yes and no. AI systems can already be trained to perform very specific tasks and outperform humans, and this could also be the case for some types of art which are more associated with ‘manual’ techniques, for example digital images and digital paintings in a specific style. We are probably not there yet in terms of manually painting a physical painting, which would be more a question of robotics (in which I am not an expert), and not AI. However, while it is clear how AI can outperform human experts in, say, detecting a specific form of skin cancer or winning a game of go against a human champion, it is not even remotely clear to me what ‘outperforming an artist’ means. It surely is a subjective question. Yes, one could imagine a situation where the majority of people would judge an art piece produced by AI to be superior to a number of similar pieces by humans, but I don’t know whether this matters at all.
How will artificial intelligence influence artistic intelligence?
Henrik Junklewitz: If by that question you mean ‘will artists be inspired or led by AI tools to craft novel and interesting art pieces?’, I would say that I am convinced we are going to see many intriguing new forms of art influenced by AI. If you mean ‘to what extent will AI be able to develop artificial forms of artistic intelligence, like inventing a new art medium?’, I would be consistent with my previous answers and say that it is very unlikely to happen anytime soon.
Does science need art? If so, why?
Henrik Junklewitz: I think science does not need art to function or make progress towards its intrinsic goals, per se. However, I do believe that humans fundamentally need art, and will always engage in artistic activities – to be creative, to relax, to communicate, to find new ways to look at the world, or to inspire their mind to go beyond its everyday boundaries. Ultimately, it’s humans who do science. And I do believe that art also has a profound influence on the creativity and freedom of mind of scientists. It certainly does on mine.