Depreciating the Real: The Never-Ending Disappointment of AI

Knowing a great deal about what is in the world through photographic images, people are frequently disappointed, surprised, unmoved when they see the real thing.

Susan Sontag, On Photography, 1977

What happens to our perception of the world when we are subjected to a systematic barrage of edited images, texts, voices and environments? The coarseness of reality begins to wilt under the effects of automated airbrushing. Memory is transmuted into fantasy, and whatever is left behind as a trace of the real, like a photograph, must be subject to the comforting distortions of editing. It is common to have a negative reaction to each new AI development, from late 2022’s ChatGPT to Dall-E, an image-generating software also developed by OpenAI. But when we are subjected to a barrage of banal reproductions of our own images, our judgement is weaned to accommodate the product as a convenient labour-saving device.

As one account put it, our intuitively negative response is a ‘prejudice [that] will fade. In just a few months, generative AI has gone from novelty to everyday tool’. It has also shifted from a labour-saving tool to a way of outsourcing emotional life, ‘personal’ communication and whole relationships. The outsourcing trialled in corporate and industrial settings is increasingly being applied to consumer markets, pushing technologies and programs designed to automate marketing and branding ‘ever deeper into our internal nature’, as Guy Rundle writes.

This shift has been heralded both here and elsewhere as ‘the real deal’, marking a need to take seriously the effects of once-rudimentary AI technologies on almost every aspect of life. Their integration into everyday consumer products makes them nearly impossible to avoid, effectively enlisting swathes of people into contributing to building AI datasets that will, in a perverse feedback loop, be used to ‘enhance’ their lives. AI is installed as a seemingly necessary appendage to any new piece of software or device, and tech corporations automatically integrate AI into almost all functions. This reflects AI’s general tendency to bypass intention, deliberation and reflection. We ‘forget’ about the background role these technologies play, not because they become irrelevant or outdated as some propose, but because they are taken for granted.

In her seminal book On Photography, Susan Sontag challenges the view that images steal reality. Instead, she argues, reality finds it harder and harder to measure up to the images that saturate our perceptual field. She writes that images ‘tend to subtract feelings from something we experience at first hand and the feelings they do arouse are, largely, not those we have in real life’. With the steady incursion of AI images into consumer image-making, our feeling for reality loses ground even further. We must face a bleak disappointment with reality that, at its worst, becomes a sort of nihilistic wish for the end of the world, already rampant in climate and accelerationist discourse.

The withered real

With AI image platforms, obstinate reality is substituted for an ontology of convenience. Sontag quotes Ludwig Feuerbach’s 1841 The Essence of Christianity, defining ‘our era’ as one that ‘prefers the image to the thing’. This despairing tone is echoed through photography’s multiple transformations of the image. As Sontag reflects on these transformations, she notes her contemporaries ‘responding to an increasingly depleted sense of reality.’ Like other arts before it, photography, having threatened the depiction of reality, now found itself facing an impoverishment of material and technique with which to continue to stake its claim on authentic representation.

Criticising AI image production is not simple. Representation is no longer really an aim. For a long time now, certain commentators have spoken about photography as ‘capturing’, ‘storing’ and thoroughly mediating our experience of reality. Sontag recognised this potential and more, insofar as ‘technology has made photography an incomparable tool for deciphering behaviour, predicting it, and interfering with it’. She anticipated what more recent writers such as Brett Frischmann and Evan Selinger, in Re-Engineering Humanity (2018), describe as the way in which ‘tools become part of the environment that shapes our beliefs, preferences and capabilities’. Their concern is that our accommodation to tools like Google Maps does not simply allow us to achieve existing goals (like navigation) more easily; instead, it shapes and redirects our goals, working both cognitively and materially to reshape the world.

The recent flourishing of AI software, introduced as a convenient and desirable consumer product whose real implications remain temporarily obscured, is another link in this chain. Part of what we need to do in order to bring its implications into focus is disaggregate what is ordinary, like the activity of editing, from what is not, such as the automaticity with which images, texts and voices are subject to editing configured by prefabricated systems that determine what counts as the proper subject of representation.

Rachel O’Dwyer, in Tokens (2023), analyses the acceleration of techno-dystopian fantasies driven by an increasing despair and pessimism about our world. Dissatisfaction with economic systems drives crypto internet communities. They seek (flimsy, reactionary) shelter from genuine social problems. Online games offer the feeling of what psychologist Karl Groos called ‘pleasure at being the cause’, of which we are starved by our jobs and political disaffiliation. O’Dwyer describes how technological fantasies are put to work in the real world, shilled and upvoted ‘through meme-hustling’ so that the ‘dream of the metaverse is that someday the rendered world, with its virtual things and virtual money, will feel more real than the world of things that can’t be called from the command line’.

One problem is that the fantasy of escaping a disappointing reality by leaping into virtual reality encounters exactly the same disappointment at the banality of the ‘metaverse’. So too may image editing ‘improve’, in some limited sense, the shine and ease with which we view our lives. But we remain disappointed, not simply at the depthlessness and blandness of the images created by the AI but at the reality to which we must return. AI editing erodes both our ability to tolerate a rough-hewn real world and our ability to imagine another one, better and yet equally real.

On photography’s ‘assault on reality’

The novelty of what many call ‘AI’ should not conceal the history of the endangered status of the real. Provocatively arguing that images ‘make an assault on reality—which is perceived as recalcitrant, as only deceptively available, as unreal’, Sontag anticipates the momentous changes wrought by the photographic image. Affecting first painting and other pictorial arts, the mass availability and almost-instant development of photographs fundamentally changed our relationship with images. She writes,

Photographs do more than redefine the stuff of ordinary experience (people, things, events, whatever we see—albeit differently, often inattentively—with natural vision) and add vast amounts of material that we never see at all. Reality as such is redefined … The photographic exploration and duplication of the world fragments continuities and feeds the pieces into an interminable dossier, thereby providing possibilities of control that could not even be dreamed of under the earlier system of recording information: writing.

To disaggregate any genuine novelty in AI from an immanent development of technology, we should keep in mind the development of image-analysing techniques, which goes at least as far back as the mid-nineteenth century to what Jonathan Crary calls Gustav Fechner’s ‘epochal formalisation of perceptual experiences’ as measurable data. Fechner established ‘measurable units of sensation, quantifiable increments that would allow human perception to be made calculable and productive’. This should alert us to the fact that AI software ‘sees’ in code. Fechner’s fantasy of measurable perception disenchants vision—if, as AI engineers do, such a metaphor can be taken literally.

As a number of observers have written, Google’s data-gathering strategy has shifted from pure monetisation through advertising to a more explicit collection, storage and analysis function. This supports its AI aspirations, as Viktor Luckerson points out:

The ability to quickly categorize people, places, and things is the entire selling point of Google Photos, of course, and facial recognition helps achieve that aim. But as Google’s AI techniques become more sophisticated, the company is weaving an ever-growing web of relational data about the world.

In order to seamlessly introduce AI that predicts and serves consumer preferences, Google must ‘quickly categorise people, places and things’, developing ‘an ever-growing web of relational data about the world’. Google is open about this, quoting Rick Osterloh, SVP of Google Devices and Services on its store page: ‘Your phone needs to understand your world—your context, your unique needs and preferences, how you speak, and what you care about’. This data serves the aim of selling back to consumers a personalised corporate fantasy, which paradoxically ends in an ever-and-always-the-same product, or what Sontag calls the depersonalised narcissism of the photograph, as well as subtly modifying consumer preferences in line with their imperialistic business model.

Everyday life was transformed by the image-taking and image-making capacity granted by cameras. It is re-transformed by the data-taking and data-making activities of online life. AI image editing makes the distinction between the ‘taking’ and the ‘making’ increasingly hard to sustain. The photographic image is a manipulation—a chemical or digital process mediating the effect of light on some form of receptor. AI constitutes a break not because it manipulates images but because it does so in a way that is concealed from the image-taker. Allison Johnson, a tech reviewer for The Verge, embraces the shift from ‘photo’ to ‘memory’ because ‘Memories are elastic and imperfect. They’re subject to our biases and moods, and they change over time. Generative AI is about to be everywhere, and people will want to use these tools to make their photos look more like their memories…’ The recording and potentially corrective functions of the photograph, which re-orient us to a past whose instant we have caught, are lost to the slippery enhancement and distortion of fantasy.

On editing for convenience

Far from being able to interrupt the continuous stream of consumptive fantasy, edited images adapt themselves to it seamlessly. As Paul Virilio writes in The Vision Machine (1994), ‘they are preparing the way for the automation of perception, for the innovation of artificial vision, delegating the analysis of objective reality to a machine…’ Delegation makes the peculiarly hyper-real quality of professional photographers available to a deskilled audience that has lost interest in the image’s depth. There is a thrill in seeming to be able to produce professional images without recognising that a skill is being lost and under-valued. Tailored for social media environments in which what counts is a distracted glance, our eyes scanning rather than looking, AI editing makes changes that ‘nobody would see … at a glance’ as it erases entire people from pictures.

Available, cheap, convenient, techno-scientific innovations become ‘addictive’, according to Joseph Weizenbaum, and ‘create a concrete reality, a self-fulfilling nightmare’. Google offers to not only ‘make’ and enhance your photographs but, by ‘organising them into an engaging sequence’ and storing them, turn them into a confirmation-screen against the intrusion of reality. Google’s latest ‘Pixel’ phone has a tool called ‘Magic Eraser’ that cuts out unwanted distractions, from people to frowns. But Johnson’s caution about the potential ‘alarming implications’ of image-editing AI are quickly dismissed once the market-friendly ‘family’ is introduced. She writes, ‘it’s also the very camera feature that every parent would love to have, especially those with more than one kid … Being able to bring it all together in one image—one shining moment that technically never happened—is the dream. Right?’ The move from potential ‘evil-doers’ to automatically making your child appear to smile is accomplished in the space of an innocent rhetorical question.

On consumption and disappointment

The image-editing affordances of the Google phone turn reality into a shopping experience, in which we choose the most immediately gratifying representation from a stock list of preselected options. Designed to organise the record of our lives along consumptive lines, AI image editing also appears to be targeted at sellers. One study, co-authored by someone quoted in Google advertising material, demonstrates the ability to manipulate interior design features in a way seemingly perfect to enable real estate agents to hawk dreary hovels as light-filled mansions. Architecture critic Kate Wagner has already noticed a trend towards blandness on real estate apps like Zillow, saying that just as fashion images ‘distort our perceptions of our own bodies, staging distorts our idea of what home is and what a good house looks like … [They] pathologise what is, in reality, perfectly fine’. ‘Fine’ is intolerable in a world made available to pure fantasy. Since, as Wagner writes, ‘we don’t often see the interiors of real houses and the messy truth of how most of us live except when we visit other people’, but instead see celebrity interiors mocked up by the same dull commercial designers for faux-candid video tours, we become inured to both the fantasy and the reality.

‘Through photographs’, wrote Sontag, we have ‘a consumer’s relation to events’—so much so that the distinction between experiences that occurred to someone else and to me is blurred by ‘habit-forming consumership’. Photographs decontextualise an event, representing the experience through a chosen image. As Google and other tech businesses take on the work of establishing and defining context for us—figuratively stealing it from the images via automated organising and storage systems—the scope of ‘choice’ in image-making narrows. Art critic Ben Davis writes, ‘AI-powered customization tends to cause people who use it to isolate themselves in symbolic systems that are more and more difficult to communicate to anyone outside of their bubble, causing sociopathy’; this may sound rather dramatic, but there is a kind of sociopathy in the mundane process of editing out entire groups of people from, say, a holiday beach scene, as contemplated by the advertising material for Google Pixel. The social world is rubbed clean of strangers, every landscape rendered sublime by the solitude of the chosen figures. Every package deal holiday is turned into an adventure for just you, and you dream a fantasy in which the world is just for you. The image makes it seem possible. And reality cannot but disappoint you.

Never fear, for it is unlikely you or anyone else will do much more than glance at one of the pictures in the unquantifiable scale of image production (3.2 billion images shared online daily, for example). Instead, as Daniel Palmer writes in ‘Photography as Indexical Data’,

photographs taken by humans in the old-fashioned manual way are increasingly ‘viewed’—or ‘read’—by computers, most often to enhance, classify or categorize them … Humans never set eyes on a lot of these images, but every single one of them is analyzed by software algorithms. Computerized vision was designed to automate tasks that humans would otherwise do, but the reality is that only computers are capable of ‘viewing’ the gigantic aggregations of images contained in today’s databases.

For Virilio, this machine-image made for machines is ‘an enigma’, the seeing of which occurs through ‘“vision machines” designed to see and foresee in our place’. This foresight is, of course, essential. AI editing does not simply ‘see’ the image, but ‘foresees’ your preferences, or what it imagines them to be—or what Google might like them to be—and offers a convenient pathway to achieving those consumer preferences. As Johnson admits, ‘I can see myself using it to take distracting objects out of the background of a scene or remove a wayward booger from my kid’s face’. In a similar article, she repeats the claim: ‘Personally, I have no problem with the ethics of swapping out an image of my child’s face where he’s blinking for a nearly identical one where he’s not’. This seemingly innocent gesture evinces disappointment at the child’s (real) booger-ridden, blinking face. If only the child would conform a bit more, says the image, feeding back the fantasy via its virtual achievement.

On fantasy mistaken for memory

Freud theorised screen memories as a defence against real memory, the remnants of which are repressed in the ‘falsified memory’, as he writes in the 1899 paper ‘Screen Memories’. Similarly, Sontag describes how

images are indeed able to usurp reality because first of all a photograph is not only an image (as a painting is an image), an interpretation of the real; it is also a trace, something directly stencilled off the real, like a footprint or a death mask … a photograph is never less than the registering of an emanation (light waves reflected by objects).

The real, for Freud and Sontag, is buried underneath whatever distortions are later applied. It leaves a trace like a smear, a smudge—an irrefutable blur immune to the crisp illusion of our fantasy. Yet with ‘Photo Unblur’, Google will sieve your old photographs and sharpen the images, smoothing skin tone and foregrounding a face as you wish you had seen it rather than as you saw it. The mystery of our memories is less solved than it is destroyed by the overlay of fantasy.

The shift from ‘photos’ to ‘memories’, along with the advertising claim that AI photo-editing is ‘allowing you to easily enhance your memories’, signals a movement from the manipulation of perception to the intervention in inner life. As the outer collapses into the weakened inner, breaching our defences against immersion in pure fantasy, so too the distinction between ‘taking’ and ‘making’ images collapses. Photographs do not ‘turn out’ (or not) anymore. Instead, they are made after the fact to have ‘turned out’, whether reality was recalcitrant or not.

As identical photographs accumulate on platforms and ‘cloud’ services, Google promises to allow you to fantasise that your identical picture of the beach is really unique to you. With the overlaying of its standardised editing, your memories of being bored, tired, hot and hungry are gently falsified. Sontag warned us not to imagine that the image was taking on the role of fetish and becoming real, but to watch for the way reality was becoming shaped by images. Our choices, consumption, expectations and behaviours come under the purview of image-making tools; our lives are tailored to service an image-processing industry that feeds back into the tools we, innocently, use.

Having seen a majestic Ansel Adams photograph or an immense Sebastião Salgado landscape bathed in an immaculate high-contrast stream of sanctifying light (or the easily edited Instagram imitation of it), you go in search of the experience offered by the image. When you arrive, the light is not quite right, and the perspective the image promised seems impossible. Still, you only have five minutes, so you snap a photo. But as Allison Johnson asks, ‘Do you want to remember the dull lighting you saw when you got to the Grand Canyon? Or do you want it to exist in your memory—and your photo library—more vibrantly?’

What, exactly, did you visit: the image or the place? Your disappointment at the promised land(scape) has been muted by the ability to edit in your lighting preferences. You have turned the world into a studio. The world disappoints by failing to conform to our apparent preferences, so we ignore it and reshape our perception along the lines of a fantasy, using tools that afford certain changes. When we lose the landscape irrevocably, ‘finished, nearly finished … Grain upon grain … a little heap, the impossible heap’, as Clov ponders in Beckett’s Endgame, the image will haunt us as a registry of the fantasy to which we succumbed.

The degraded world

Neal Stephenson’s 1992 parody dystopia Snow Crash has become a template for almost every monopoly tech company, including ‘Meta’, Apple, Amazon and Microsoft. In it, people escape their dissatisfying physical world by ‘disappearing in the Metaverse’. Lost, inevitably, in the process was the irony of Stephenson’s novel, but one line lives on as a meme: ‘when you live in a shithole, there’s always the Metaverse’. The trouble is that, as Rachel O’Dwyer writes, ‘The shit in the metaverse is spilling over into the shit in the real world’. The virtual’s bland reproduction of corporate vision, overlayed with the promise of absolutely unique personalisation, cascades onto whatever is malleable in the world.

Boring, recycled and uniform real estate ads take up and dominate mental real estate with boring, recycled fantasies. Our commitment to being in and seeing the world wavers. The so-called ‘Internet King’ Marc Andreessen, a venture capitalist, even chides us with the accusation that living in ‘a real world-environment’ is a privilege, and that ‘Everyone else, the vast majority of humanity, lacks Reality Privilege—their online world is, or will be, immeasurably richer and more fulfilling than most of the physical and social environment around them in the quote-unquote real world’. Another tech writer, Beau Cronin, writes that ‘Many parts of real life, it turns out, have been poorly implemented’. The real world is full of snags and inconveniences that interrupt the comfortable perception that everything is made for me: ‘Reality has had 5,000 years to get good, and is clearly woefully lacking’, complains Andreessen. As Talbot Brewer writes, ‘Fantasy inures one to the world, seizing it as an infinitely pliable tool for private distortions. The workings of fantasy … produce nothing that is alien to the self, there is no effective resistance to their impetus’.

In this world of depersonalised narcissism shaped by a uniform corporate directive and the imperatives of comfort, convenience and consumption, it would be quite sensible to lose trust in reality—and not simply because someone is trying to deceive us in some grand ideological scheme (that might be available for critique), but because no one is, or no person at least. The image-making occurs apart from any human desire to play, to lie, to distort or—extraordinarily—even to take pleasure in an image. With reality bypassed, the image also begins to lose its allure. Disappointment rubs off. It is very easy to disillusion—to erode a common world for the sake of convivence or expedience. It is much harder, and more painful, to bring back the grainy texture, to remind ourselves of the time it takes to do things together. What we take for mere tools often have destructive effects. We do not make it easy for ourselves to get the world in view.

Meta-Facebook: The Quest for the Infinite Office

Timothy Erik Ström, 30 Oct 2021

Zuckerberg describes the metaverse as ‘an embodied internet’. In this, he could not be more wrong: the metaverse is precisely about deepening the forces of disembodiment.

About the author

Scott Robinson

Scott Robinson is a writer and academic with work published in Overland, Arena, Memo Review, Index Journal and elsewhere. He is a former editor of demos journal and associate editor of Philosophy, Politics, Critique. His website is found here.

More articles by Scott Robinson

Categorised: Arena Online


Comments closed

Support Arena

Independent publications and critical thought are more important than ever. Arena has never relied on or received government funding. It has sustained its activities largely through the voluntary work and funding provided by editors and supporters. If Arena is to continue and to expand its readership, we need your support to do it.