When German artist Mario Klingemann started using computers to “augment” his creativity, the worldwide web and Google did not exist. Over the years, the German artist had to tweak the new computerized tools he was being offered, writing his own plugins for tools like Photoshop or After Effects, before 21st century technologies allowed him to build more complex systems for algorithmic generative art.
|The Butcher’s Son by Mario Klingemann|
Today, Klingemann who creates artworks using algorithms, code and neural network, is considered one of the pioneers in the use of computer learning in the art. The Google Arts and Culture resident recently won the 2018 Lumen Prize for digital art for The Butcher’s Son, a sitting nude image that had been entirely generated by a machine using a chain of GANs (generative adversarial neural) networks. First developed in 2014, GANs are a class of artificial intelligence algorithms used in unsupervised machine learning, with two neural networks contesting each other in a zero-sum game framework.
“These algorithms are called “adversarial” because there are two sides to them: One generates random images; the other has been taught, via the input, how to judge these images and deem which best align with the input,” explains Ahmed Elgammal, director of the Art & AI lab at Rutgers University.
In the case of The Butcher’s Son, Klingemann built a system from multiple neural networks that explores the human form. “It does that by first generating random stick figures after having learned the possible variations in human posture by analysing hundred thousands of photos. Each of these figures is then passed to a GAN that has learned to translate them into a rough sketch that reminds of a painting - interestingly that painterly look is the result of the model making mistakes and abstractions, since I did not train this model on painting, but actually on pornographic photos. In the final step this sketch is given to another GAN which tries to fill in the missing details and adds interesting textures and artefacts,” Klingemann explains.
The final winning portrait, with its disfigured face and blurred flesh color tones, was reminiscent of Francis Bacon’s artworks, in line with a distinct GANs aesthetic that reflects how the algorithms process information.
“Visually there are typically surreal distortions in localized areas which create a very interesting part-whole relation in the painting: what seems like a typical beach landscape might suddenly in one region sprout a forest,” explains Karthik Kalyanaraman, one half of the curatorial collective known as 64/1 which focuses on art for the post-human age and which curated last year a show entirely dedicated at AI art at Nature Morte, one of the largest contemporary art galleries in India.
Kalyanaraman believes an art expert can usually identify work produced by a GAN because of tell-tale textural effects, “a kind of shimmer or a scrambling of various image-making techniques like impasto, dry paint, etc,” though artists are now working on overcoming this ‘GAN signature.’
Elgammal, for example, has developed a system he calls AICAN — a ‘creative’ rather a ‘generative’ network, that is specifically programmed to produce novelty. The program, which can works on his own, was essentially created a “creative collaborator and partner” for artists, and some of these collaborations with Tim Bengel and Devin Gharakhanian were premiered at the fair SCOPE Miami Beach in December. “We designed AICAN with the intention of opening up exciting possibilities for artists to explore new territories of their own creativity and artistic process,” said Ahmed Elgammal.
Before its appearance at SCOPE, AI-generated art had already been in the spotlight with a controversial sale at Christie’s New York, which saw an AI artwork smashing its $7,000 to $10,000 pre-sale estimate to hammer at $432,500. The portrait, titled Edmond Belamy, had been created by GANs trained by Obvious, a Paris-based collective that signed the work with the algorithm name instead of their own, which seems to indicate that the algorithm was the real artist.
Who is the artist?
So are AI machine about to replace artists? Not so, say art practitioners.
“The narrative that the AI is the ‘artist’ is absurd and that is clear once one starts to understand these algorithms. Humans conceive the algorithm, teach the algorithm a particular visual style by curating the ‘training set,’ and uses their aesthetic eye to curate the final output! In so far as there is not a shred of autonomy or will in the working of this process, I think it is really premature to call the AI an artist. Do we think the lens and the camera and ‘nature’ which provides the setting is the true artist and not the photographer?,” says Kalyanaraman.
“A GAN by itself is often just an empty vessel,” concurs Kligemann, adding “as an artist you have to fill it with content by deciding what to train it on and then finding and curating thousands of images that you want to extract some essence from.”
“Machines can probably only ever be as creative as humans, not much more, since if they came up with radically creative ideas we would not be able to recognize their brilliance since we would simply not be able to understand it. That is the problem with human imagination - it can only expand slightly beyond our horizon - new ideas take time to be accepted and understood,” Klingemann remarks, “But like any other technology that allows us to do things faster, better or with less effort AI already seeps into our daily life and will not stop at helping us with our imagination and creativity. Just like you probably can't imagine a life without your mobile phone anymore in the future people might not be able to imagine how people in our times could have had ideas by themselves without the help of their creative assistant.”