BCA Talk | Mario Klingemann: Art needs context and certain rituals to be fully appreciated
Mario Klingemann is a leading pioneer in the AI art movement. His featured work for the “Virtual Niche” exhibition, 79530 Self Portraits, on loan from Qinwen explores how a feedback loop of GANs is transforming errors into meanings.
Mario, can you tell us a bit about your background?
I am 51 years old which probably means that I am one of the senior citizens of the current scene of computer-generated art and in particular the NFT art world. I am part of the very first generation of children that grew up with digital technology, starting with digital watches, pocket calculators and the first handheld games like “Game & Watch” by Nintendo and a few years later with the advent of the first home computers in the 1980s. I taught myself programming when I was about 12 years old and found that this was a form or creative expression that felt the most natural to me and is a way of working that aligns well with my personality. I always was interested in image-making, but painting did not really work for me, so I was more drawn to photography. I had my own darkroom and tried to create images that I could not capture with a camera alone which is a motivation that has stayed with me until today. So over the years, I co-evolved with technology and at some point, I could use the computer exclusively to explore the space of “all possible images”. I am entirely self-taught since somehow I was always too early in the things that interested me and those were not taught at art school or universities yet. A few years ago the time had finally arrived that it became possible to create images with the help of deep learning or “AI” as it is commonly called and since this was a field I had always had been interested in, but which was not feasible or accessible until recently, I was ready and prepared and could dive right into it and explore all the new possibilities. Practicing digital art for so many years, there was always one aspect that turned out to be the hardest: making a real living from one’s art. Selling digital art has always been harder than selling paintings or sculptures since many collectors still prefer physical artifacts. I was lucky that at some point I got enough recognition in this field that I got invited to international art shows and festivals and could sell my works through my gallery. But as we know now in the past year the revolution that is NFTs swept through this world gives digital art the value that it deserves and also allows to present and sell it in its “native” form without having to use a physical vessel.
Can you talk us through the process of using neural networks and creating “79530 Self Portraits”?
79530 Self Portraits is created through a feedback loop of generative adversarial neural networks (GANs) that transforms errors into meaning and reveals biases in the training data. I had trained the GANs for this work on classic portraiture from western art history. One of the models in this system has learned to transform an image into a very simplified biometric “sketch” that reduces the input to just a few lines that mark the facial features the model recognises. The second model in this chain has learned to reverse that process: it gets such a sketch and tries to reconstruct a “real” image. What I do is to couple these two models together in an infinite loop, so each model receives the output of its counterpart as its input. What happens is that since both models are not perfect and make “mistakes” the data that circulates through the system is constantly transformed and altered, but at the same time not entirely random. It could be seen as a little mountain river that does have a recognizable bed that meanders around steps and whirlpools but the water that runs through it is never really the same.
The reason this work is also a self portrait is that I introduced a camera into the system that gets mixed in in a very small quantity with the data stream. This allowed me to influence the flow of the information and add another source of “error” and control that the models try to reinterpret. The number 79530 is the number of frames – every single one different from the other one – that this video consists of and it hints at the fact that these neural models unlike humans can keep on creating new information or images forever, without having to sleep or taking a break.
You are considered a leading pioneer in algorithmic art and generative art. In your opinion, how does computer art differ from traditional art?
For me, the biggest difference is that in traditional art artists usually try to perfect a single work at a time whereas in algorithmic art you try to create and perfect a system that can produce an infinite amount of works within a given possibility space that the artist has created. Of course, it is also possible and legitimate for a generative artist to only select singular outputs from a system they created and present those as the final piece, but personally I find the “making of a world” aspect more fascinating and interesting. You are faced with the problem of trying to predict the future behavior of a system you made: will it stay interesting, relevant or surprising? Can you trust that there are no hidden errors in there that will make it break or stop at some point? Did you manage to explore the possibilities and parameters of the world you created deep enough or did you overlook some treasures that are hidden in some hard to reach sub-spaces?
Do you think AI art has the power to convey emotions and how?
Do you think that books or movies have the power to convey emotions? I think we can agree that this is so. Stories can convey emotions even though they might be invented and not “real”. AI art is to some extent a subcategory of storytelling – mostly purely visual so far – and whilst we are still in an early phase and have not yet explored all of its possibilities I have no doubts that AI art, which ultimately is just sourcing from our collective human experience and transforming it, will be able to tell us stories that we can relate to and that will convey emotions. In particular, since AI art is still created by human artists who put their thoughts and feelings into their work.
When machine learning can produce art on its own, how can we measure human parameters in artistic creation in your opinion?
I would say that it is the other way around: the moment we believe that we can measure or detect “humanity” in artwork and that work has been produced by a machine on its own, that is the moment when that machine has created art. The difficulty lies in the definition of what “on its own” means. So far none of the AI that we are working with is fully autonomous, they are still steered and controlled by human artists or engineers. These machines and models are like levers that require us to exert very little force to generate very complex results, but it is still us who pull that lever.
What are your views on the current NFT bull market?
I think in all its excesses and hype it is a very important step. Digital art has been around for more than 50 years and is a vital part of the way we all live today where “digital” is an essential part of everything we do, but until recently it had not been given the recognition and appreciation that it deserved in my eyes. This is the first time in history that digital artists can make a living purely from their art and do not have to sell their talents in other ways like making commercial work or having to use their talents in better paid but unrelated digital technologies.
I see the current excitement and exaggerations as a natural phenomenon that you can observe in any complex system into which you inject a new “agent” or new information. When training my neural models I can observe similar patterns of volatility and “battles” for the right of interpretation which come in waves that fluctuate a lot and go to undesirable extremes until they find a new calmer equilibrium that satisfies the needs of all parties.
The exhibition “Virtual Niche” is the first and largest offline exhibition for NFT/crypto-art, which will be held at UCCA Lab, Beijing and at Jin Art Space, Shanghai. What do you think of having a physical space exhibiting digital artworks?
I think physical exhibitions and dedicated spaces for those are a vital part of our culture. Art needs context and certain rituals to be fully appreciated. And that is what these spaces provide. They allow the visitors to focus their attention and also provide the actual space, scale and atmosphere that cannot be provided by a mobile device or by sitting at home on a sofa and looking at a wall-mounted tv.
