[Asha De Lanerolle]
Text 4844/Image 10
234876—23 LDN
FEAR THE MACHINE
When AI image generators first appeared on the scene, my conversations were dominated by the seemingly urgent question of whether these machines would ever be able to generate ‘real art’ or truly ‘realistic’ images. I would spend the tail end of nights out talking with friends about the intricacies of human creativity, trying to untangle these from the computational language that we have come to use interchangeably to describe human thoughts. We were fascinated by the weird conformity of the AI art appearing on the internet, and amused by its more obvious failings — the hands with too many fingers or Chat GPT’s inability to do even the most basic maths. I tried to make sense of this web of thoughts, and in what became the starting point for this piece, realised that I held a deep and primordial fear of this emergent technology. This fear was partly rooted in an ideological opposition to tech-billionaire transhumanism and the surveillance apparatus of big tech. But beyond this, the prospect of a faithful image conjured up by a machine-learning algorithm felt simply unsettling, somehow unholy.
In the midst of the extensive and warranted controversies that surround the development of AI, our all-too human experience of this technological shift often falls out of focus. I wanted to reckon with this fear, interrogate my reactions to this ‘unholy’ but sometimes strangely beautiful AI content, understand its emergent aesthetic and think about what it might mean for our experiences of the digital worlds we all inhabit.
Playing around with Open AI’s Dall-E 3 model, I initially felt somewhat vindicated in my scepticism. As hard as I tried, I could never quite get Dall-E to produce anything but strangely rendered and overly fantastical imitations that could straddle— but never quite surpass— the uncanny valley. Prompting these models is an unfamiliar experience in itself, as you navigate the often counterintuitive associations that these art-bots devise. Ask for a ‘realistic image’ and you would often get a less realistic result. Although when you think about it, you wouldn’t necessarily flag a photograph as ‘realistic’, while a slightly uncanny CG render would likely get such a label.
On one occasion I asked Dalle-E for an image of an abandoned building, which returned a decently realistic but overly-staged result. When I asked it for a ‘not rendered’ look, I was instead gifted an image with a pair of hands in the corner, holding a rifle. When I asked it to remove the gun, it simply could not see it. And stranger still— its revised version, with ‘no guns or weapons’, now contained a whole battery of anti-aircraft guns instead.
While sometimes funny and questionable, these generations carry a sort of hauntological record. The ghosts of the data-set beneath the resulting image. Sometimes, in the more obvious slips, these spectres become tangible and traceable. The vaguely post-apocalyptic figure of an abandoned building, combined with the word ‘rendered’, likely aimed Dalle-E’s gaze at the video-game scenes and concept art it was trained on. The gun, generated in the same POV you might find in a first person shooter, was therefore right at home in this composition, not understood as a separate object with a vastly different meaning, but as a common thread in this aesthetic domain. These slips also produce an alluring visual strangeness, and as I will explore later, it is here where I think their most interesting creative potential might lie.
This process felt deeply conflicting, like I was sampling the forbidden fruit of creative expression. And, despite the fact I could make images I actually liked, these productions felt empty. The people and places within them, while sometimes deeply life-like, had no histories or assertions of their own. Like the character art that fills much of their training sets, they are figures whose story can only exist in external world-building that is not innate to the image itself. Aside from the text in their prompt, they capture no real context or history. But even though I knew all this, when staring at the faces of these algorithmic hallucinations, I couldn’t help but start to assign them narratives and offer them names.
The most realistic images left me questioning not just the potential effects of this technology, but also my own relationship to real photographic images and the meanings I find within them. They unsettled the last notion of documentary truth which we often assign to photography, one already eroded by photoshop and CGI. They shift the way we understand the images that have become so integral to our contextualisation and experiences of our surroundings and each other. The mere awareness that they exist, momentarily renders every real photograph a potential AI hallucination until established otherwise. And, the fact that they can be conjured with just a few lines of text invokes a less subtle crisis for the value of human creative work.
Such a crisis in both meaning making and creative production is not, however, new or necessarily unique to AI. The invention of photography itself once posed a similarly biblically-perceived threat to traditional painting, as image generation now does to the photograph. Writing almost a century ago, Walter Benjamin chronicled the early resistance to photography captured in what he calls a ‘chauvinist rag, the Leipzig City Advertiser’ which argued that: “To try to catch transient reflected images...is not merely something that is impossible,... the very desire to do so is blasphemy”. Reading this, it was difficult to ignore how
the act of blasphemy felt like a pretty good analogue of how I felt using Midjourney.
Like the painters of that time, we too are left to reckon with the emergence of a new medium of representation and an ease of reproduction that has never before existed. Before trying these models, it was simple to try and ignore this, gesture to the derivative creations of the ‘prompt engineers’ and disparage it all. But within our collective disdain for AI art itself, there lies an aspect of potentially naïve self-soothing as we grapple with our uncertain futures and a threat to the practices and value of human creativity.
Benjamin was interested in the effects of the reproduction of art in modernity and the way in which this would inherently degrade their aura — a slippery and somewhat vaguely defined term which captures an objects presence and history; its ritually defined authenticity and uniqueness as an original; and a state of distanced interrelation or the ‘experience of an object gazing back at one’. His ideas are both immensely interesting but can also be frustratingly inconclusive, for he sometimes laments the destruction of meaning and authenticity imposed by modernity and reproduction, whilst arguing for the emancipatory potential of this destruction, which would free art from its ritualised and institutionalised roles in society.
In many ways this is still the unresolved paradox we are still wrestling with today. I too can see the theoretical benefit of this emancipation, yet I am captivated by the aura of great art and photography, and unsettled by its cheap recreation through the hallucinations of an AI. Far from resembling a revolutionary overhaul of art, the generally representative AI image is somehow deeply commercial. Often appearing lifeless, they often invoke the inoffensive genericism of a stock photo despite their typically bright and expressive content. The prevalence of these kinds of images may in fact obscure the latent capabilities or uses of these models. But, in attempting to decipher why they so often look like that, we are offered an insight into the true nature of the technology and, perhaps, popular art.
The early adopters of these models were not preoccupied with the aesthetic logic and hierarchy of the capital A Art world, instead influencing our perception of what ‘AI art’ looks like through their embrace of fantasy, character and concept art that seldom graces a gallery wall. The widely-used ability to instantly copy any artists’ work has been particularly damaging to many upcoming and established creatives. But at the same time, who people choose to copy has also greatly influenced our view of AI art. In the year Midjourney first went public, the most popular artist to mimic was not a well-known name like Da Vinci or Van Gogh but a relatively obscure fantasy artist named Greg Rutkowski. (McCormack et al. 2024). So despite the volume and breadth of content the models are trained on, many of their images emerge with an almost uniform artistic style. This style reflects not just the tendencies of their user base but also the major aesthetics of the art in its training data - scraped without permission from online repositories like ArtStation and Deviant Art.
This ‘AI look’ we have come to dismiss is therefore an unflattering but not totally inaccurate mirror of what a lot of digital art already looks like. Rather than challenging the art logic of the establishment, it mostly competes with art that exists largely in parallel to, not in competition with the more traditional art world. And, even on these platforms, glimpses of revolution emerge not out of an embrace of AI art, but in mass opposition and visual protests against its display and the AI partnerships their owners have entered into.
Benjamin may have underestimated the extent to which our hierarchies and conceptions of ‘aura’ would evolve around and intertwine with the inherent reproducibility of photography — and subsequently almost anything digital — rather than be annihilated by it. But he grasped a truth about photography’s relation to its subject and reality which illustrates how the aura can endure reproduction. Describing an early photograph of a woman taken by David Hill in the mid 19th Century, he observes that there is an obscure yet tangible trace of the individual that allows her to transcend mere depiction in the image, and ‘never yield herself up entirely into art’. Even in the more intricately staged and, to him, less interesting photography of the time, the camera captures some innate authenticity within the moment that often overrides, even if in only small ways, any attempt at staging and control.
AI images, even in their general derivative conformity, carry their own traces, not of our perceived reality, but of our images and representations of it. No matter how obscure the prompt, or alien the output, some trace of the data-set remains. Much has been written about the ways in which these models perpetuate the harmful biases of these data-sets and the demographics they are trained by. But to acknowledge these biases in visual models, in combination with the way they are optimised to please the widest audience possible, is also to acknowledge that their outputs may therefore offer an ugly but important reflection of our popular visual culture; its most problematic and reductive tendencies included. As Alimma Aldiyar observes whilst exploring AI’s stereotyped depictions of Buddhist deities, AI models may be ‘the tool that has the strongest grasp of which images are holding control over our cultural consciousness’. They reveal the undercurrents in the images and art we both make and consume. And as we question their assumptions, we are forced to examine the individual and collective associations from which they emerged.
Benjamin regarded Hill, and later Eugène Atget — who is now viewed as one of the originators of what we might call ‘street photography’— as revolutionary because they approached photography free from the self-importance of Art. Many of Hill’s contemporaries attempted to mimic the depictive logic of painting at the time, including the fake columns and the staged poses employed to allow subjects to stay still for a long painting session— despite the speed of a photograph. But Hill’s photographs were simple and unadorned, often ironically taken as references for the paintings to which he ascribed much more importance. Meanwhile Atget’s now influential images of bleak pavement scenes and fading corners were peddled to little fanfare as street-market souvenirs. But, by presenting these empty scenes under titles like ‘Paris’, usually reserved for photographs of famous landmarks in excessive and staged splendour, Benjamin argues he ‘initiated the liberation of the object from the aura’, breaking the spell of these cities and our associations of them.
I have a fascination with these places. These are the lands and images of Terrain Vague. The ruins, unlikely corners and third spaces which dot parts of our cities and to which we have come to assign a profound and sometimes problematic sense of authenticity. The Spanish Architect Ignasi de Solà-Morales attempted to characterise our photographic obsession with these interstitial places, often found in the former or current industrial lands, abandoned buildings, urban infrastructure and run-down neighbourhoods. To de Solà-Morales, our experience of architecture and our urban environments are always intertwined with our visual depictions of them, ‘our gaze has been constructed and our imagination shaped by photography’ (de Solà-Morales, 1995). Atget’s images of mundanity and decay may have disrupted the perceived grandeur of the city, but in doing so these places and their images took on their own aura and influence in our Psyche.
The first time I used Dall-E and Midjourney I could not help but ask for images that might invoke these places — abandoned factories and concrete ruins. In the feigned realism of the generations I could partly conjure the ‘territorial indications of strangeness itself’ that characterise the photography of terrain vague. Though when all went to plan, I could not quite capture the feeling one gets seeing or walking through these grimly beautiful voids. And yet the eccentricities and failings of the generation — the seemingly impossible physics, the strange overlaps, the uncannily incorrect biology, the merging of adjacent objects and the creeping-in of strange motifs like the spectral gun — effected their own language of ‘strangeness itself’. To me these are the most interesting possibilities of image generation. For it is in the uncanny opposites to ‘realism’ where the machine excels. In manufacturing weirdness; unsettling but sometimes beautiful movement and morphing; or in randomness and ethereal decay, difficult to manifest with a human hand or eye.
Where this strangeness becomes embodied with what can be read as an artistic purpose, an intention and direction beyond a simple prompt, I think we find AI art that feels at home with what we would traditionally regard as genuine creativity. In Marianna Simnett’s visuals for the work Firefly we encounter the unsettling trance-like degeneration of the artist’s face and performance, surrounded by amorphous butterfly wings. Stitched together from 1000s of images from a generative adversarial network (GAN) model, Simnett describes the process of producing this work far differently from what we might think AI assisted art to be. It involves constant wrestling with the model, integrated with a great deal of human work and iteration in order to achieve the shots and consistency that are congruent with her intention.
Whereas in Jon Rafman’s music video for a version of Kanye West’s Vultures [Havoc edition], we are transported to an ever-morphing nightmare of bleak imagery. As if the early memes made with Dall-E mini — with melting faces, 7 finger hands, and figures or body parts that seem to vaporise and reappear, were re-awakened through an arcane ritual. The music video is a head-on embrace of AI’s strangeness and ‘faults’ but here all these mistakes are the point, they become its atmosphere. It is a compilation of the scary offcuts that a quest for realism would have long thrown out, and in it we become hypnotised.
A new music video produced for “The Hardest Part” has been touted as the first commercial music video made with OpenAI’s Sora model. While its overall aesthetics are rather unremarkable, Its director, Paul Trillo, describes a similar wish to embrace ‘the space between canny and uncanny’ that could be produced by the model. It seems I am not alone in my enjoyment of the strangeness produced by AI. And, in some ways these visual eccentricities have become a sort of medium in themselves. Distorting the relation between the subject and realism that the camera ushered in, through their imperfect embrace of the uncanny valley between the two.
To me, the question of whether AI can produce ‘real art’ is therefore already answered. But the ramifications of this are not experienced as a singular moment, where side-by-side in a museum an AI image is anointed as equal to some famous old painting or influential photograph. As with photography before, the emergence of AI creates a broad shift in the mediums through which we experience our world and its spaces, which today exist both on and offline. Thankfully the AI eclipse of human creativity has not yet been realised. But it has already invaded every facet of our digital world. It is here where I think we venture into un-navigated territory and where our worst incentives align. It is also here where the world seems more and more surreal by the day.
Why cyberspace feels like moving through an ever-expanding brain-rot meme as we grapple with our deeply violent present and possible futures, while wading through a torrential flood of inhuman nonsense that has infested our feeds.
The endless void of the scroll is for many of us as much of a space we inhabit as the street we might live on, the local supermarket, or our favourite night out. As with the real world, it is in these spaces where we socialise, find community, share and expand our thoughts or simply wander aimlessly. I think many of us have become so used to spending our time in online spaces that we cease to consider them as such. Much like the local grocery store, we simply learn its aisles and wander, mostly thoughtlessly, sometimes blocking out the world with our headphones on, until we find some combination of nourishment that might make a meal. When I open Instagram, I do something similar— not truly taking in each post as they merge into one another and the oblivion of the scroll. It is only when something really jars us out of that half daze, that we can truly take scope of our online surroundings, and our position within them. Every day the content might be different, but the experience of the scroll in its totality remains largely the same.
The wilder hobbyist internet of the 90s is long gone, but in the gentrified fiefdoms of big tech there are still voids and eccentricities that remain, the terrain vague of the digital world. The moments where the invisible walls seem to melt and you are algorithmically transported to an online community, discourse or level of terminal online-ism you wouldn’t have imagined existed and could never have searched for. In many ways, this is why I engage online; the sheer eccentricity of humanity and our propensity to form community from our shared interests, experiences or humour, no matter how obscure. It is in these moments that facets of internet culture and their increasingly porous online-offline boundaries are spread and subverted. And, where we are exposed to the lives and political struggles of those both similar and vastly different from ourselves.
But in doing so we are simultaneously everywhere all at once in both time and place, unable to singularly focus. The move away from chronological feeds has furthered this temporal compression. The past becomes breaking news, the present becomes the past and the future melds into all. We are decoupled from our territorial base, globalised yet never standing still long enough to plant our feet and explore. This is a wildness in its own way. A difficult to articulate yet common textural experience that we share despite the hugely different content on our individual feeds.
The photographs and videos which are so integral to our online experience have until now, tethered us to at least some semblance of time, place and glimmers of personal realities, no matter how constructed. Already, these images and the words or voices of their creators have become ‘content’. Artists and makers to journalists and politicians can no longer just share an extension of their ideas, work or selves. Instead they must curate a personally branded melding of the three to build up a para-social fan base. Images and art have become increasingly moulded by their role in the production of content, rather than as cultural artefacts in themselves. As a commenter on a process art video rather profoundly observed: The instagram reel, not the pendulum paint bucket, was the artist’s true medium. And in the medium of the mindless reel, TikTok or post, there is no artist more prolific nor more suited to exploit the hidden intricacies of the algorithms than AI.
Regardless of their future trajectory, the combined powers of LLMs, image and video generators, and AI voiceover, have already enabled a never-ending mass of meaningless content and fake interactions to engulf our feeds. I actually enjoyed the deep fakes when they weren’t quite perfect — absurdist and deeply human concoctions of brain-rot and anti-capitalist irony that could only exist in this very moment. Like @Prognozpogodi69’s videos of Steve Harvey reading out corporate waffle from an industrial glycine manufacturer, accompanied by no real attempt to lip sync his superimposed figure. But now I question almost every clip I see. I was used to the monotone of TikTok’s AI voice over. But these days Chat GPT sounds like Scarlett Johansson in Her (2013). A case of sci-fi turned so real that only seems to demonstrate just how out of touch the people who captain this technology probably are. Even the memes now have AI generated captions about seemingly random topics, an uninterpretable move in a bot vs bot chess game between spammers and ‘the algorithm’. Whether we consciously realise it or not, the textural experience of our online worlds is undergoing a profound and possibly irreversible shift.
But amidst the noisy hype around AI, it’s easy to forget that it has only been a few years since a similar hype emerged around cryptocurrencies, the blockchain NFTs and the metaverse, whose pundits promised would revolutionise our digital world. Once the initial hype subsided, the components of so called ‘Web 3’ emerged as not just as an overly ambitious technological failure, but also a moment defined by a collection of massive scams, enabled by blind faith in techno-futurist con-men.
The NFT craze which emerged as part of this became a very public exposure of procedurally generated and largely terrible ‘Art’ designed to appeal to teenage hype-beasts, internet celebrities and their follower bases. This ‘art’, exemplified in the widely derided Bored Ape Yacht Club, was in some ways an interesting precursor to discussions about AI art and the paradoxes of how technologies are promoted compared to their material impact. The seemingly progressive rhetoric for a decentralised, freer and more accessible web of the future, juxtaposed with a reality of rampant cons and speculative investing was both jarring yet simultaneously a true moment of our times. A melting pot of consumerist masculinity, Californian ideology and old fashioned grift which resulted in a strange confluence of hype-beasts, tech enthusiasts and future finance bros following their deified shills to either make it big or, more likely, lose it all.
But before the chaos and opportunism — which also left masses of unsuspecting and often vulnerable investors in financial ruin — there was perhaps a glimpse of genuine interest in a decentralised cyberspace that could be conceptualised as something beyond a dollar line that would keep going up. Or— in the case of NFTs— art freed from the veiled confines of its established institutions and gate-keepers as Benjamin might have hoped. A true ‘metaverse’, built on human creativity and interaction might have harkened back to the eccentricities of the early web. The one we got, exemplified in projects like Decentraland, have become what video essayist Dan Olson described as a ‘Dead mall’. A largely abandoned and eerie environment dominated by crypto advertising, poorly executed brand collaborations and legally grey gambling.
It is in the shadow of these digital ruins that the AI revolution has come to the fore. In the realm of the physical, ruins can be spaces of possibility. Places of terrain vague, where attempts to impose singular use and order are subverted by time and reclaimed by people and nature. But taking a virtual walk through Decentraland, the ruins of the digital feel more like the back-rooms than an overgrown temple or an underground squat rave. They are spaces which seem to capture the experience of hyper-capitalist monotony, inhabited only by the eternal unconscious of bots. As AI powered content continues to proliferate, the internet and social media are starting to feel more and more like that dead mall. A place full of dead possibilities and unreciprocated connections. Like crypto before it, the greatest impact of AI may be the sheer volume of its most mediocre outputs, peddled by bot accounts across the internet, rather than its highest artistic achievements.
As Maggie Appleton points out, to avoid this we may retreat to the gated safety of a smaller and smaller web, shielding ourselves with greater and great ironic cynicism. But as AI models continue to outsmart our means to detect them, limiting the number of bots may involve more and more stringent and potentially biometric identity verification. Something which would have far reaching effects on the generally more anonymous centres of internet culture like 4Chan and the depths of queer Twitter (now X) or Tumblr, who’s strange combination of opposites serve as the vanguard for the production of internet culture.
There is a possibility that some of the emerging AI technology will follow a similar fate to that of Web 3. It certainly appeals greatly to much of the same fan base and its current hype-train shares the often-unquestioned assumption that exponential growth is not just possible, but essentially pre-determined. It is however a more generalised and likely less fundamentally limited technology than the blockchain, so maybe only time will tell how powerful it might become. Regardless, the same juxtapositions in rhetoric and reality remain. It is impossible to seriously discuss AI while ignoring the problematic context of its creation, use and monetisation. While pundits and, rather surreally, the founders and leaders of key AI companies have been openly fear-mongering about the potentially existential threat posed by a true artificial general intelligence, or AGI, this tends to obfuscate our understanding and critique of the already-existing impacts as those same founders expand their reach. The essential mass theft of art and intellectual property which was scraped to train these models, their massive energy and water consumption, and the threats posed to jobs and livelihoods.
It is also impossible to understand AI solely through its more benign use cases, such as image generation, when its development is deeply embedded in the military-technological complexes of the world’s powers and the increasingly intertwined surveillance apparatus of states and big tech. AI models can offer politically convenient, and falsely ‘objective’ authorities, which can be used to enforce censorship, decide employment prospects, and even impose carceral justice. And most disconcertingly, versions of AI technology which can make you a summary or a postcard are also being developed and used as weaponised surveillance tools, or to determine the targets of genocidal assassination in Gaza. Painting with a grenade would be an overt political choice. And, I think we must ask whether the sheer dominance of ‘big tech’ can continue to shield their tools from the same characterisation.
Trying to make art with Dall-E often felt like painting with that veritable grenade. But in its flaws and strange beauty, AI imagery reveals much about our own gazes. It offers a chance to reflect on our societal relationship to images and how we understand the people on both sides of the camera. It forces us to reckon with the sometimes un-inspiring and questionable under-currents of our visual culture. As creativity becomes content, the true arbiters of that culture may no longer be those who hold sway in established cultural institutions, or on DeviantArt, but rather the murky algorithms of the social media platforms.
The images of the camera came to define and express the prowess and problems of the modern world and its projects. Their partially enduring aura’s continuing to shape our cultural understandings even amidst their widespread reproduction. Maybe, in a sick twist of fate, the AI images which contribute so readily to the banal absurdity of the moment might be one of the mediums which best invoke our experiences of it. In their uncanny visuals we find the enduring ghosts of surveillance. Their data-sets born from stolen amalgamations of online life and creativity, assembled into algorithmic approximations which express our very human prejudices. The AI image, whether beautiful or boring, carries the most prescient contradictions and conditions of our lives in the architectures of ‘Surveillance capitalism’ (Zuboff, 2018). The same architecture which made its very creation possible.
In the haunted expanse of the internet, where our every experience or desire becomes data used to trap us in an endless loop of mindless consumption… maybe those ghosts are more at home than we wish to admit.
- Text/Asha De Lanerolle/Sam Harding
- Designer/Pauline Hill
❶ [09] img9