As our perceived world continues to be changed and challenged through technological advancements, many artists and non-artists alike agree that artists fulfill a unique role in society that technology will never be able to replace. However, considering that technology is a product of the human mind, it is not inconceivable that humans could one day translate what it means to be human into an algorithm that a computer can understand. Engineers and artists like Gene Kogan are attempting to accomplish a task similar to this by creating deep-dreaming artificial intelligence networks. These networks mimic the human brain and are able to output bizarre collaged images through image recognition and comparison algorithms.
In Invisible Cities, Gene Kogan continues to explore the possibilities of creative human and computer collaboration. The project runs on a neural network that is fed images and forms relations between the images it knows and new images that it continues to be fed. A neural network is a stack of processors in which each processor performs a unique task, analyzing a problem from multiple perspectives and suggesting possible actions. This process mimics the way that neurons in our own brains communicate with one another, look at things from different angles and create connections between what we know. Work in this field attempts to bring us closer to understanding our own creative processes.
Gene Kogan has written several artificial intelligence algorithms. He worked on the project, Invisible Cities, along with several other members of Open Dot studios during a workshop he led. The network was fed around 500 satellite images of cities and trained to recognize the different components that make up these cities. This allowed the network to be fed any image and translate it into a satellite view map resembling a city that is compared to that image. This becomes an application for existing cities to be mashed together, or for a user to import a drawing or photo that is not a satellite view of a real city. The network will turn drawings into a satellite views of a city to output a new quasi-city. There are several models of the AI which transform images they are fed into the styles of various cities, such as Milan and Venice. The idea for this project came from a 1972 Italo Calvino book exploring imaginary and mystical cities of the Mongol empire dreamt up by Kublai Khan and Marco Polo. These maps are imaginative speculations on the desirable constructs of cities and the many possibilities that they hold.
The maps that the network generates exist somewhere between the physical and digital realm. There are notions of a computer thinking in abstract nature. The algorithm creates an image that is not quite CGI but not quite a believable arial map. These cityscapes seem to be dreamt up or at least abstracted notions of cities that we as humans are accustomed to. There is a sense of a subconscious aspect to these images akin to surrealist paintings. This is contradictory to the fact that the images are synthesized by computers which have no real thoughts at all rather than what they are told to do. Compared to other AI or “deep dream” art created by neural networks, the Invisible Cities AI creates compositions that feel more tangible and more justified. The Invisible Cities AI seems to have intent in the way it creates artificial landscapes collaboratively imagined between man and machine.
The success of Invisible Cities stems from the project's equal reliance on both human and artificial thinking to create imagined landscapes. This brings forth the conversation of what computers can and should be used for. It is exhibited here that computers and neural networks can not only come up with connections between images that are not intuitive to humans but can also do so much quicker than our brains can and produce tangible results. As suggested by Kogan himself, both of these traits are incredibly important in numerous fields as we move forward.
Kogans work is on the intellectual side of art spectrum. It results from a combination of research, engineering and creative applications. Many of his works could be interpreted as experiments with artificial intelligence software. In another project, Deep-dream and Dense-cap, Kogan explores applications of multiple networks working in parallel in an attempt to understand each other. The result is a video showing an infinite zoom created by taking single frames and running deep-dreaming algorithms on each frame. Another network analyzes the images and attempts to recognize what real world object they are. The network places a colored box over sections of the video displaying the names of the images it can recognize, such as nose of dog, eye of bird, cup of orange juice, etc. In comparison to Invisible Cities, Deep-dream and Dense-cap, is a less artistic approach to machine learning. However, these two works are similar in that they create speculative scenes on future applications or artificial intelligence.
Being an emerging technology, artificial intelligence inspires artists to experiment, leading to very different results. Fuck, a video work by Turk Lees explores algorithms similar to the ones Kogan experiments with. However, in a major conceptual leap, Lees decided to feed the algorithm the most abundant content on the internet, pornography. The results are scenes that appear vaguely sexual, extremely cluttered, and undoubtably surreal. The work is essentially porn scenes composed of heavily mashed together images. Patterns, eyes and animals heads are some of the more recognizable forms. In comparison to Kogan’s work this piece conceptually delves into a different space. Invisible Cities creates surreal scenes of cities. However, cities are constructed concept, and something that cannot occupy the subconscious mind. Sex and porn are common themes in surrealism, and occupy the subconscious. Considering this, it is interesting to imagine what what could occupy a computers subconscious. The result, much like Kogan’s works, are things that humans could never imagine or create themselves.
These pieces represent what can be achieved when algorithms are thought of not only as a means to quickly solve a problem but also as a way to quickly output a pattern recognized by an algorithm that does not really solve anything. Computers lend themselves to alternate forms of thinking, far different than the way our own brains operate. Studying these algorithms can lead to a greater understanding of how our brains may operate. Generative artists have been exploring this notion for years. In this way, artists have been collaborating with AI for multiple decades and the notion that computers and artists cannot create art collaboratively has been dismissed.
Gene’s Website: http://genekogan.com/
Article by Motherboard by Madison Margolin https://motherboard.vice.com/en_us/article/techie-artists-used-machine-learning-to-turn-hand-drawn-sketches-into-city-maps
Article on Fuck by Turk Lees