Describing works by talking about the technique with which they are made coincides with an academic approach in many cases necessary. It helps the cataloguing and the insertion in thematic tanks with which we are used to tell the history of art. In a discourse of definitions, technique over time has given way to concept and abstraction: the abstraction of art before the manifestation of art. Today, one of the most experimental fields is the digital one. The metaverse pursued by Mark Zuckerberg and Meta is not just a commercial possibility, but a space for creation. NFTs are auctioned off like Flemish oils on canvas and the language is no longer that of great painting technique or conceptual art, but a language of computer codes, as in the case of GAN Art.
GAN (generative adversarial network) consists of a set of neutral networks that are trained through an algorithm to generate new data. From this data set the algorithm derives information and creates new images or videos, creating shapes but also hyper defined objects, which have an important feature: they don’t really exist.
The possibilities of creating works increase and change in relation to the information stored by the network, and the role of the artist changes accordingly. The GAN artists are positioned closer to the concept that the Renaissance had of the profession, they are computer craftsmen, in almost all cases without an academic training in favor of a numerical one. GAN Art is open to improvisation, understood as a creative result, as spontaneity, but not in the moment of creation. Handling codes and algorithms necessarily requires computer know-how. Images of subjects that do not exist are opening up numerous possibilities for gallery owners and auction houses, with deviations and points of contact with more contemporary expressions such as memes. For the first time a moment is being created in which art is produced and no longer reproduced.
What is GAN?
A few years ago, the page “This Person Does Not Exist” (created by Uber software engineer Phillip Wang), which generated faces of non-existent people through GAN, made a lot of noise. The page was intended to “raise awareness” as stated by Wang, and open the door to the infinite possibilities of GAN and not trivialize the discourse to examples of deepfake.
Ian Goodfellow himself, the first to introduce this technology after studying computer science at Stanford in Deep Learning and Machine Learning, thought of GAN to “improve the state of the art in terms of quality and distribution.”
Just click on the website thisxdoesnotexist.com to understand how distressing it can be to look at GAN Art’s hyper-realistic images. Abstract images, in which colors are mixed without defining precise shapes, have a different effect. In the latter there is one of the most interesting aspects of the art form, namely the impression of incompleteness. There are gaps in the definition of the image and curiosity arises from that feeling of catching vaguely familiar or already seen shapes, simply because it is so. It is possible to combine in a single image or video all the works of the same artistic current, or with the same theme, as done by Anna Riedler together with the programmer David Pfau in Bloemenveiling (2019), a series of tulips taken from 17th century paintings.
As mentioned, GAN Art does not open many avenues to improvisation; the examples cited so far and those that will follow are accumulated by a necessarily computer-based approach before an artistic one. Even the potential continuous reworking of the works, one of the great possibilities of the medium (think of a video game in which the landscape is created endlessly, always different) requires an enormous amount of work to correct the codes and re-educate the algorithm.
GAN Art does not have a limit of reception, so much so that lately it is increasingly exploited for events and installations. Bulgari recently set up a sensory experience in Piazza Duomo in Milan, while Christie’s already talked about Artificial Intelligence and GAN in its Art and Tech Summit in 2019.
Artwork that does not exist
At thisxdoesnotexist.com you can not only recreate artwork such as paintings, but also craft and design objects, antique amphorae or chairs, as well as other more bizarre categories such as ideas, satire or emotions. Even song lyrics, maps, words, eyes or horses, all created in machine learning.
Mario Klingermann, a pioneering artist and programmer of GAN Art, has worked with various realities, curators and musicians, among them the collaboration with Massive Attack for their EP Eutopia is an example. Klingermann is also the first artist to have seen one of his Artificial Intelligence works sold at an auction house. It was Sotheby’s that hammered out the work “Memories of Passerby” in 2019 for £40,000.
Other interesting projects are that of Noah Veltman who generates movie posters (fun trying to recognize them), or that of Kazakh Amir Zhussupov who created an algorithm capable of combining the illustrations of Hans Ruedi Giger in a single work. Sofia Crespo chooses to focus her research on the forms of nature, Robbie Barrat on nude images, which complicate even more the discourse on censorship and the inaccuracy of the algorithm that defines what is right or wrong to publish on social networks.
As mentioned, all these artists are primarily technical. In this way a provocative friction is created, certainly not intentionally, between the figure of the technical nerd who sees images as numbers and the traditional version of the visionary artist, accustomed to abstraction.
This role of GAN is not secondary. The dialogue and the questioning of the identity and role of artists has coincided in many cases with moments of great innovation and rupture. The growing interest on the part of the art world for NFT and AI, positions GAN Art’s works as one of the most contemporary products with the greatest expressive potential. It is able to exploit more direct languages such as memes, receptive, made of symbols, texts and layered layers, which create a similarity with the information contained in the overlaid data.
With the development of GAN in different areas of culture and information, a problem of protection could arise, which has already been encountered for example with deepfake videos. GAN can already create fake news, fake job ads, fake real estate ads, fake resumes, and songs that no musician has ever played. It remains a territory to explore to reason about the origin of artistic expression, the participatory role that artists have in a community, even a digital one, and how cuddly those cats are that, unfortunately, don’t exist.