The Aura in the Age of AI
An analysis of the presence of AI-generated visuals in the art world, using Walter Benjamin's cultural criticism as a framework.
In his seminal essay “The Work of Art in the Age of Its Technological Reproducibility”, Walter Benjamin theorizes on revolutionary media in the post-industrial age. With the rise of new mediums comes seemingly endless debate surrounding what defines art. Century-old talking points continue to be raised amidst the steadily increasing presence of artificial intelligence in the art world and visual culture at large. Using Benjamin’s thinking as a framework, it can be observed that AI technology as an art medium is following a similar trajectory to that of print, photography, and film, in that it is taking the degradation of the aura to new heights.
The majority of AI-generated artworks are created using generative adversarial networks, or GANs, a two-sided algorithm. The user enters a text prompt, one side generates images, then the other side determines whether or not the image matches the prompt. This determination is based on the training set, a massive quantity of images that the programmers fed to the program, which it then analyzed. The determining side then rejects unsuccessful images, and the generative side tries again; this process is repeated until a coherent image emerges. The output relies on the content of the training set and the sophistication of the GAN, which improves with time. These programs do an imperfect job while in their infancy; the earliest publicly available GANs produced images that would seldom trick the user into thinking they were looking at a photograph. Results varied from creepy and grotesque to beautiful and painterly. Google’s DeepDream consistently incorporated canine forms into its outputs, even when the prompt entered had nothing to do with dogs, simply due to the excess of dog photos within the training set. While failing their tasks, the algorithms dreamed up more fascinating imagery than what they create today, through their firm understanding of how things look but not how they operate. This confusion amongst the earnest attempt to create an accurate image is humanizing. It is counterintuitive that the result of an algorithm following directions is so spontaneous. Combined with the artwork’s intangibility, this strange, unpredictable quality has the power to instill a genuine sense of wonder in the viewer, a feeling that resembles Benjamin’s concept of the aura.
The aura refers to the “here and now” of an artwork, a tie to a specific time and place. It is associated with the act of creation that elevates it beyond the presence of a simple object and elicits a response from the viewer. The aura of an artwork has historically granted it its value and utility. Benjamin argues that, in reproducing the artwork through mechanical means, the aura is diminished. In the case of an artwork produced by a GAN, there is no “unique existence in a particular place”. It was created inside an algorithm, its first and likely only home is within the digital realm; the thing to marvel at is the fact that it was created by a computer. An AI-generated artwork living inside the blockchain only exists in the form of data, there is no object, it lacks both a time and place of creation and an original. Interestingly, it is singularly influenced by what already exists, but it in itself is unique. These works are not reproductions, while they are informed by vast datasets of visual culture, ultimately they are new images. Reality reflected back to us by a machine; a distilled essence of whatever prompt was given to the GAN. Within the context of Benjamin’s framework of the power of images, these artworks lack a true aura, but there is something else there. A kind of simulacra that did not exist before. Returning to early AI-generated visuals, there is a genuine auratic quality in the early iterations, that has now been lost as these consumer-friendly programs have become the skilled imitators that they were designed to be. However, this whimsy can be recaptured by making the process more human.
Many contemporary artists yield fascinating results when thoughtfully incorporating machine learning into their practices. Harold Cohen was one of the first to work with algorithms as a generative art practice. His machine, named “Aaron”, created drawings with a distinctly childlike look to them, with time becoming more refined and technically skilled. Cohen treated the machine as an extension of himself, referring to it as his “other half” and giving it equal credit in the creative process. Helena Sarin is a prominent contemporary artist who works with AI. She uses GANs trained on drawings that she has done by hand. This approach is conversational, a creative collaboration between human and machine. Another notable recent example is Refik Anadol. Using the Museum of Modern Art’s collection as a training set, Anadol’s programs interpret the feverish artistic progress of the 20th century into their own visuals. The resulting body of work, Unsupervised, was shown at the MoMA and has been lauded as a display of machine creativity, an attribute usually only used when describing humans. It is an interesting juxtaposition of motives, in that modernism was all about deconstructing form and innovating new styles of art, and GANs can only create based on what already exists. This body of work is an exciting look into how the mechanical imagination works, only cheapened by the fact that it is up for sale on the blockchain. Making these artworks into NFTs feels like an attempt to conjure up an aura; doing this aligns it with historical tradition in the sense that an artwork is granted authenticity in having an owner. It is hard to ignore this desire to capitalize off of revolutionary work. Rather than being a noble foray into the philosophical world of machine creativity, Unsupervised becomes a money grab.
The blockchain is not the only place where the sale of these artworks occurs. A Christie’s auction in 2019 saw a major sale of a portrait created by a GAN, fetching $432,000. The piece has been compared to the distorted faces of Francis Bacon, although there is significant difference in that the AI’s portrait is conceptually hollow, it did not create the distortion intentionally for artistic value, the algorithm was simply imperfect. The fact that it sold for such a massive sum questions the notion of an artist’s intention being what qualifies art as art. In the same vein, the contemporary art world has taken into consideration the identity of the artist, generally speaking. The presence of the individual maker, a celebrity-like concern, elevates certain works to incredibly expensive heights. The lack of authorial presence in AI artwork is unique; this novelty is an undeniable addition to the enthusiasm. With the price tags being fetched for these artworks, AI has settled comfortably into the market.
As can be plainly observed with the medium of film, new media with great potential for reproduction is an excellent source of capital. Benjamin took an optimistic stance on the revolutionary political potential of film, in its decisive departure from artistic ritual, but as time has gone on, the movie industry has only expanded into more and more of a commodity. With such a wide reach, the narratives of the most mainstream movies must be as universal as possible. There is an adherence to ritual for the sake of mass appeal and therefore the end products are politically inert. Benjamin’s contemporaries who made up the Frankfurt School posited the concept of the culture industry, the machine of mass media being produced and broadcasted for the sake of keeping the masses entertained without drawing attention to the freedom that they lack. The public gets to participate in the arts without a prerequisite of intellect while the industry is able to make more entertainment and profit.
Thanks to AI, the process of creation has become simpler than ever; the clunky human involvement has been edited out for a more streamlined experience, and the results are predictably uninspiring. Cinema and machine intelligence are combined in the interactive multimedia experience HyperCinema, a venture out of New Zealand. Participants have their likenesses placed into a fictional narrative through technology that superimposes their faces onto the characters. Within the content of the story itself, it is a classic hero’s journey containing shameless references to the Marvel Cinematic Universe. The experience is ephemeral in its personalization, the viewer gets to indulge in seeing themselves on screen, if only for a couple hours. The visuals generated by the AI sit at the cusp of the uncanny valley. The viewer sees themselves thoroughly blended with the virtual world; the results are discernible, but the novelty and immersive nature of this experience seem to counteract the unease. Endeavours such as this one make no attempt to reintegrate the aura. Prophetically, Catherine Russell writes about the postmodern cinematic experience in reference to the aura, “New media has not reinvented cinema as an auratic object but as a complex and multifaceted form of experience.” observing that in the 21st century, video reigns supreme, at the expense of its most human elements. It is reasonable to believe that it is only a matter of time before the culture industry starts pulling generative technology out of its toolkit at every turn.
It is unsurprising that there is this lack of truly revolutionary art being made by AI, given its adherence to existing narratives. Social inequalities as well as biases in visual culture have led GANs to generate images that reinforce cultural stereotypes. When a training set contains degrading representations of certain groups, these representations will proliferate. GANs are entirely objective by nature, they are trained to recognize and recreate therefore they cannot make the distinctions that a human artist can. They lack sensitivity. Western culture dominates the internet, which is the training ground for these programs, so naturally they reinforce this culture. Groups that are underrepresented in the art historical canon are less likely to appear in a painting generated by a GAN. Additionally, there has been widespread anger about the methods that tech companies have used to train their software. In order to produce images that are polished and successful in fulfilling the prompt, AI has to analyze a massive training set. Large companies with access to large swaths of data have used an unknowable but definitely massive amount of artwork for this purpose without the artists’ consent or any form of compensation. The lack of consideration towards human artists points to profit being the primary concern when it comes to AI-generated artwork.
Regardless of the fascinating contents of AI art, the feelings of unease that many artists get from this medium should not be ignored. Marshall McLuhan famously stated that “the medium is the message”. Following this line of thinking, one can infer that in giving machines the ability to make art, an act that for eons was only done by humans, humans are being equated with machines. If the medium chosen to create a work of art barely requires any human effort in order to create a human-esque result, the message is that it does not matter who creates a work of art or why they create it. All that matters is that the image exists and it can be looked at. Sanctity is lost. Returning to the idea of the aura, with its presence or degradation, Benjamin places immense importance on the experience of viewing an artwork, what we get out of it before analyzing its formal and conceptual qualities. An artwork that is so profoundly ungrounded in human perception can feel insulting when treated exactly like something made by human hands. Additionally, the speed at which GANs create images is impossible to keep up with. The “here and now” is becoming saturated with the uncharming kitsch being churned out by the mature programs. In the world of AI art, much is missing, the human touch, the profound catharsis of human expression, the aura; these things are not needed when it comes to generating banal imagery for the sake of capitalistic endeavours.
The context within which AI art arose is inextricable from the medium itself. Entirely detached from tradition and yet completely enslaved to what already exists, there is little revolutionary potential. What more can the GANs do, beyond improving their parroting? As the visual culture of our modern era grows more inundated with images with each passing second, the aura is fainter than ever.
Bibliography