Strange Shadows [Part II]
Two centuries of photography have prepared the way for algorithmically-generated images. Now the image-world is becoming a Museum of Babel.
Welcome to Ever Not Quite—essays about technology and humanism.
This is the second and final part of a two-part essay, the first of which came out last Friday. Together, these two parts constitute one essay. If you haven’t already done so, you’ll probably want to read Part I before you proceed.
Next Friday—and for no other reason than that I want to share some of the most interesting sources I’ve come across on these themes—I’ll post a curated collection of quotations about photography, images, and the visual sense which have informed the background of this essay and contributed to my thinking, bringing this theme to a close for now.
As always, thank you for reading. If you find value in the newsletter, please share it with others—it’s the best way for new readers to discover it.
These observations, elaborated and digressed at much greater length, were explored in a series of six essays written in the early- to mid-1970s by Susan Sontag—perhaps the most sensitive exegete photography ever had—which were collectively published in 1977 under the title On Photography. By that time the camera was no longer new, and the activity of taking pictures no longer considered strange. The history of photography was littered with scores of celebrated practitioners, moribund schools and ideological movements whose influence had already risen and fallen, and innumerable declarations proclaiming what would be the camera’s true legacy—not to mention photographs, millions of photographs, which suffused the vision of all who partook of the modern media sensorium. All this provided the twilight backdrop for reflection, for taking stock of all that had changed in the image-world since the 1820s. Now, the images of things—once the laborious offspring of accomplished artists, often folded into practices of ritual observance—had traveled far along the road of emancipation, their autonomy relative to real things more completely realized than ever before. Their eventual digitization still decades from becoming commonplace, photographs were material, usually paper reproductions made out of the very stuff of the world whose image they endlessly multiplied.
Writing in her accustomed mode as an omnivorous critic of modern culture, everywhere attuned to metaphors and their consequences, Sontag concluded the last of these essays, “The Image-World”, with the following appraisal:
The powers of photography have in effect de-Platonized our understanding of reality, making it less and less plausible to reflect upon our experience according to the distinction between images and things, between copies and originals. It suited Plato’s derogatory attitude toward images to liken them to shadows—transitory, minimally informative, immaterial, impotent co-presences of the real things which cast them. But the force of photographic images comes from their being material realities in their own right, richly informative deposits left in the wake of whatever emitted them, potent means for turning the tables on reality—for turning it into a shadow. Images are more real than anyone could have supposed.1
All photographs are emissaries from a former reality. But although printed photographs provide a window onto a past which may or may not continue to resemble the present, they are also tangible pieces of the material substrate of the world itself: as physical objects in their own right, they are to be counted among the inventory of things whose images they reproduce. If reality could once claim that title in no small part on the basis of its stability and permanence relative to the fugitive being of images, the very materiality of photographs—by no means exempt but capable of being sheltered from the hazards that accompany all physical existence—enables the photograph to outlast the fleeting moment and to fix the flow of time, securing it against its inexorable annihilation.
Their unprecedented fidelity to the optical facticity of things, their unparalleled level of detail, and their mechanical reproducibility allow photographs to deflect accusations, as no painting ever plausibly could, that their portrayals represent at best distortions or reductions of reality—and perhaps even outright deceits. These features, all byproducts of the technical procedures that produce photographs, had demonstrated that the set of philosophical questions that the existence of images had always raised no longer fell neatly along lines of their truth or falsity relative to things themselves. The questions which these images posed for life within the image-world now came to orbit the dynamics of the new media environment which the proliferation of photographs had created. The photographic image-world had established a sort of parallel reality, if not more real or even more stable than the older one within which it was embedded, then certainly more compelling as an object of fascination and of desire, and as a model of perfection. This world had, in effect, colonized what had formerly been known as reality, and for the first time images could call into question the priority which it had always enjoyed: In the twentieth century, reality could be judged according to its fidelity to photographs.
With some literary indulgence, we can imagine that the image-world created by photographs might answer the question which opens Shakespeare’s Sonnet 53—“What is your substance, whereof are you made, / That millions of strange shadows on you tend?”—with an imperious confidence enjoyed by no prior class of images: I am the world’s witness but also in a sense its substitute—the record of all that is worth seeing.
The photographic universe swallowed more or less the entire corpus of every handmade image ever created. Today, most of our experiences of any given painting are, in more accurate terms, experiences of a photograph of that painting. More than simply shrinking the scale of the original painting, the photograph forces its complete recontextualization, relocating it to surroundings which the original, on account of its singular existence, could never inhabit. The next watershed in the history of photographic technology was not so much the introduction of color film, which leveled one of the most conspicuous differences between photographs and paintings and made photographs appear more like the world as we actually see it, but the transition from analog to digital—and just as photographs once swallowed hand-made images, digital media have in turn swallowed what was once a sea of paper photographs.
In a sense, even the paintings and photographs which appear on tangible surfaces have always belonged less to the material world which actually houses them than to the abstract world of images, with its semiotic matrix of cross-referencing symbols which forms what Berger once called “the great family of images”.2 And it is almost as though the back-lit images that appear on digital screens have receded more fully into this world, having ceased even to share in the light by which visible objects are seen, supplying instead their own illumination: we might print out a favorite family picture downloaded from social media for framing, or magnetically fix an internet meme onto a refrigerator door, but this is about as far as any digital image will ever stray from the placelessness of the content-stream out of which it was removed.
The rise of digital media, which made photographs more easily shareable, alterable, and reproducible further radicalized the autonomy of images—their independent existence from the things they show—at the same time that it tended to diminish their usual source of meaning. If a photograph, unaided by any caption, explains itself only by means of what is given in it—that is, what it shows is also the photograph’s answer to the question of why it shows it—then the act of taking a photograph only really succeeds in communicating meaning when it is done deliberately, and with care. The cheapness of the creation and circulation of digital images renders them considerably less likely to harbor much meaning in this sense. To the extent that we can make certain assumptions about just how little thought was involved in the creation of the photographs we typically come across online, they make no real solicitation to be taken seriously, lingered over, or contemplated. But algorithmic images extend the cheap production and surreal malleability of digital photographs, weakening almost to the breaking point our expectation that images must have any meaning at all.
The differences between paintings and the technical images produced by cameras were obvious enough that nobody would ever mistake one for the other, and provided we overlook those bizarre schools of modern painting which curiously pursue photorealistic renderings of their subjects, these images themselves disclose the class to which they belong. But algorithmic images can be made to look so similar to digital photographs that whatever distinguishes them cannot necessarily be found in what they look like, but only in the process by which they have been created in the first place. Despite their resemblance, the two are in some ways opposites: algorithmic images are produced without a lens, without any contact with the world at all; photographs on the other hand are the most worldly of images, the images with the surest tether to the real.
In a manner of speaking, algorithmic images invert the process by which photographs are produced. A photograph receives its image externally, by drawing narrowly from whatever happens to fall within the cone of light which enters the camera’s lens; but the algorithmic image is generated, as it were, internally, drawing on all the pre-existing images that have been incorporated into its training. If paintings translate the world, and photographs quote from it, then algorithmic images, however perfectly they may resemble either of these, do something else entirely: they recycle it. Or, more precisely, it isn’t the world they recycle, but earlier tokens of visual media: the photographs and paintings that existed before them. Whatever algorithmic images can be said to possess of reality comes to them only at second- or third-hand. First, paintings were swallowed by photographs, which were subsequently digitized; then, algorithms swallowed digital photographs as data to be sifted and reshuffled.
If the camera is a tool for establishing reality as it is; for documenting and preserving increasingly tenuous familial ties, disappearing cityscapes, ways of life, and ecosystems; for authenticating leisurely experiences, sights and activities enjoyed during travel, and the fun had in social gatherings; algorithmic images establish, preserve, and authenticate nothing; they simulate these things by fabricating, customizing, and remixing what older media have already captured. Paintings and photographs reveal, each in their own way, the world as the interpretation of a human image-maker, disclosing meaning as a consequence of deliberate action; algorithmic images reveal the world as a pastiche, imitation, or confection of earlier interpretations. The camera’s capacity to establish reality becomes the algorithm’s capacity to recycle photographs.
The basic proposal that is advanced by all forms of generative artificial intelligence is that, in the final analysis, reshuffled data can be used to replace the real things which had first to be converted into the data that trained the algorithm. If an algorithm performs a given function, then the artificiality by which it does so may be counted as irrelevant: if a student can be made to learn from an automated tutor, then the algorithm can be said to teach; if a machine can be made to simulate the conversation of a close friend, then it can be said to offer relationship; if human intelligence can be substituted by artificial intelligence, then it can be said to think. The strange sort of verisimilitude which algorithmic images display is derived, like photographs before them, from the fact that they are machine creations, and not the fallible artifacts of any human judgment. But while the authority of a photograph’s visual testimony proceeds from the fact that the camera was actually present for whatever it shows, algorithmic images offer a truth of a different nature altogether: trained on an unimaginably vast store of existing images, they trade in the sort of statistical validity that emerges out of a complete data set when you calculate its central tendency. This kind of statistical, or composite truth is the opposite of a photograph’s truth-in-representation, which selects and fixes a specific instant from reality.
The very essence of a painting—the fact that it is a unique physical artifact which bears the strokes of the hand that produced it—accounts for the peculiar sense in which it might be accused of being “inauthentic”. An inauthentic painting is one which has itself been falsified, a painting whose provenance as an object with a singular history—what Walter Benjamin called the artwork’s aura—has been violated by means of some artistic forgery: a “fake” painting falsifies the history of the particular work in question. When it comes to a photograph, the suggestion of its inauthenticity arises from the possibility that it may somehow have been doctored, that the reality it purports to show, or the event for which the camera was allegedly present, was not like what is shown in the photograph; the uniqueness of this particular photograph, its distinctive history as a physical object, is not at issue in the same way that it is for a painting; what matters is its fidelity to what really occurred. The doctored photograph does not counterfeit its own history so much as it misrepresents the history of the world.3
But the misuse of an algorithmic image does not elicit accusations that it has somehow been faked or doctored. There is no pretense here of meeting the standards of authenticity which belong to paintings and photographs; they make no claim to be unique objects touched directly by the hand of a great artist, nor to document the optical facticity of a scene as it really was. The distinctive malfeasance with which an algorithmic image may be charged is that of masquerading as a different kind of image than it really is—posing as or otherwise allowing itself to be taken for a painting which doesn’t exist, or more likely, a photograph of an event which never occurred. And this is due to the fact that photographs and algorithmic images are produced, as it were, from opposite directions: one through direct contact with a hyper-specific instant, the other from a thickly meditated relationship with the totality of all available data. Somehow, it seems we are satisfied with these fabrications so long as they don’t pretend to the sort of authenticity which can belong only to a different class of images.
Photography is a method for capturing, collecting, and gaining control over reality, for making it stand still as no painting ever could. In the most decisive contrast with the photographer’s restriction to reality as it offers itself, and in an odd kind of return to the most basic elements of painting—the free play of color, form, and composition—an algorithmic image is a declaration of pure ontological conjecture. No one would ever think to ask what such an image means, at least not in the way we might wonder about meaning in a painting or a photograph. These images gather sufficient human judgment and intentionality that it makes sense to ask about the greater significance of what they show: What is the event that is unfolding here, and why has the image-maker chosen to capture it? What did the author of this image intend to communicate by creating it? What judgments needed to be made in the process? These minimal grounds for finding meaning in a painting or a photograph—whether it is the likeness of a buffalo sketched on a cave wall or a provocative moment exposed onto film—are almost totally absent in an image that has been created by an algorithm.4 It is impossible to imagine paintings without painters, or even photographs without photographers. But to imagine algorithmic images created without any human action or purpose—that is curiously simple.
Sontag concluded On Photography by proposing what she called an “ecology not only of real things but of images as well”.5 Proffered only in the closing sentence of her final essay, this appeal dangled without any further elaboration. In 2003, returning to the subject of photography and the media environment which it had helped to produce, Sontag published Regarding the Pain of Others, a book-length essay which wondered about the use and abuse of images depicting human suffering: Can such images, she wondered, be employed for ultimately humane purposes? Can public revelations of the barbarities of war, of crime, or of disease, apportion visual horror in strategic or salutary ways, and mobilize the public to do more to prevent such things from recurring? Can gruesome photographs taken during wartime be used to help prevent subsequent wars from breaking out in the first place?6
Now, with the explosive proliferation of such images—indeed, of images of every conceivable sort—via the digital connectivity supplied by cable television and the internet, Sontag seemed to despair of the naïve assumptions that must underlie any program of this kind. “[T]here isn’t going to be an ecology of images”, she now conceded. “No Committee of Guardians is going to ration horror, to keep fresh its ability to shock. And the horrors themselves are not going to abate”.7 Newspapers, magazines, and TV news were not concerned with managing the psychological consequences for their audiences of the images they show, and still less did the libertarian ideals of the early internet pretend or even aspire to authoritarian programs of perceptual hygiene. No centralized authority was waiting in the wings to curate the image-world now that it had been opened up, reinvented, and greatly expanded by the camera.
Today, the image-world is larger and more autonomous than it has ever been, but what is most remarkable is not the sheer number of images daily being produced and exhibited in the infinite scroll of our digital feeds: it is that more and more of the images we see are the product of less and less thought and judgment. And this means that the image-world harbors less meaning—at least in the traditional sense—than ever before. The camera set off an explosion in the scale of the image-world, but it also forced reality to compete with images for our attention, fascination and desire, and for status as the standard by which things are known and measured; the digitization of photographs, which rendered them intangible, divorced photographs from the material reality whose image they multiply, and broke the direct optical and chemical linkage connecting the event witnessed by the camera to the resulting image appearing on the plate or on paper; and finally algorithms put the totality of the photographic inventory to work, reversing the photograph’s innate power to isolate and verify a particular moment in time, and drawing instead from the entire endowment of existing images. If there is any truth in Sontag’s pronouncement that photographs turned the tables on reality, that they had made a shadow out of the real world which once served as the measure of things, algorithmic images seem poised to drain the image-world itself of much of the very meaning which has always moved human beings to create images in the first place—but without restoring to reality any of its original dignity or status.
The allegory perhaps better suited than Plato’s famous Cave for the image-world that will be created by algorithms is a modern fable, often retold. The Library of Babel described by Jorge Luis Borges is the ultimate stockpile of textual information: every permutation of the 25 characters of an unnamed alphabet running to 410 pages. This means that the Library necessarily includes every book that could ever be written, as well as every possible error, every alternative version, and every nonsensical specimen that can exist within those generous parameters. Under these conditions of superabundance, the hermeneutic predicament is less a matter of access to those texts which expound the truth, which are freely available someplace within the library’s spectacular holdings, than it is the monumental task of painstaking excavation. Within the Library, the search for meaningful text amidst a deluge of gibberish seems to have taken the place of any motivation to produce it, and one wonders how significant such a book could be, in the unlikely event of its discovery, considering it owes its existence to mere chance rather than to wisdom. The new ecology of technical images, which might be better compared not to a library but to a museum, similarly ‘contains,’ so to speak, every possible image.8
Like writing, photography is a medium which can convey everything from the most mundane and bureaucratic facsimiles to the subtlest delicacies of genuine art. Since its invention, the camera has been used to capture images of every conceivable sort and for all manner of reasons, and algorithms have inherited this comprehensiveness from the photographic repository which they have swallowed. The Museum of Babel—the incalculable archive of all the images which these algorithms make available—thus houses every imaginable possibility: it holds every painting ever created—every Turner, Matisse, and Klimt—and every painting which could be created; it has every photograph ever taken, but also every photograph not taken; it contains a portrait of you at every age, seen from every angle; every moment: private, public, personal, historical—not to mention every alternative version. All these, but not only these, are possibilities which exist—potentially, if not actually—within this imaginary algorithmic registry, this collection of all the images which the new algorithms are capable, in principle, of producing.
This is only a metaphor, of course, and an imperfect one at that: algorithms which produce images don’t actually contain them beforehand the way a museum’s walls are already hung before they receive their first visitor, not to mention that the allegorical Museum has no curators, every image is just as likely to appear as any other, whereas the images that will populate our screens will be curated by power-brokers and influence-peddlers of all stripes. Perhaps the most consequential departure of the Museum of Babel from the Borgesian Library is that those inhabitants can’t find what they are looking for, and spend their lives in a vain search to uncover, despite the vanishing improbability of their success, the elusive texts which they seek. The images that we seek, however—as masters of the algorithm’s magic—can be found easily, conjured by the abracadabra of prompt-engineering. Defective as all such metaphors must necessarily be, the Museum of Babel illustrates two essential features of our situation as it may be shaped by the arrival of the algorithmic cascade: it is the picture of an unprecedented ecology of sorts (or, perhaps, more of a non-ecology), one in which we are granted access to all possible images—but only on the condition that they no longer embody any human craftsmanship nor possess any relationship with real events occurring outside the matrix of computer code.
The power of images to become separated from any underlying reality—which is at once suggested and also obscured by the images that reproduce it—and to establish an independent existence quite apart from the real things that caused them, is the essence of their peculiar autonomy. It was also what first enabled nature, with the aid of the camera, to reproduce itself in photographs. Photography, as the poet and essayist David Levi-Strauss has noted, was “the first manifestation of the cybernetic future of media.”9 These were the first technical images, and the first images to exploit mechanically the basic rupture which makes all representation possible. But algorithms are doing what the camera, with its dependence upon the light reflected by reality and the human being who is attuned to it, never could: they are further emancipating the image both from human selectivity and from the world itself, and causing us to discover that we have become guests, not among the shadows of a Cave, but among the images hanging in a Museum of our own making.
Susan Sontag, “The Image-World”, in On Photography, pg. 179-80.
John Berger, “Appearances”, in Understanding a Photograph, pg. 68.
I am drawing here from Sontag, who writes: “The consequences of lying have to be more central for photography than they ever can be for painting, because the flat, usually rectangular images which are photographs make a claim to be true that paintings can never make. A fake painting (one whose attribution is false) falsifies the history of art. A fake photograph (one which has been retouched or tampered with, or whose caption is false) falsifies reality.” Susan Sontag, “The Heroism of Vision”, in On Photography, pg. 86 (italics mine).
Of course, the more autonomy the algorithm had in creating the image, and the less it was guided in the process by a human image-maker, the truer this is.
Sontag, “The Image-World”, in On Photography, pg. 180
These ethical questions run through all of Sontag’s writing about photography. She had earlier written about a formative encounter she had at a young age with images from Nazi concentration camps: “One’s first encounter with the photographic inventory of ultimate horror is a kind of revelation, the prototypically modern revelation: a negative epiphany. For me, it was photographs of Bergen-Belsen and Dachau which I came across by chance in a bookstore in Santa Monica in July 1945. Nothing I have seen—in photographs or in real life—ever cut me as sharply, deeply, instantaneously. Indeed, it seems plausible to me to divide my life into two parts, before I saw those photographs (I was twelve) and after, though it was several years before I understood fully what they were about.” Susan Sontag, “In Plato’s Cave”, in On Photography, pg. 19-20.
Sontag, Regarding the Pain of Others, pg. 108.
David Levi-Strauss, Photography and Belief, pg. 63.
I’ve been interested in the relationship between photography and AI generated images for a while. They have some superficial similarities, and I’ve been trying to pinpoint why the process of taking photographs doesn’t bother me nearly as much as that of prompt engineering. You point to something important by observing that in some ways the process of photography is the inverse of that of generating an image: taking a photograph requires being in a particular place at a particular time and the result is something hyperspecific; a person can generate an AI image from anywhere and its results are less intentional/more unpredictable/in some ways more vague because they’re pulled from a repository of past images. I think you’re right that algorithmic images are “poised to drain the image-world itself of much of the very meaning which has always moved human beings to create images in the first place” and hope you’re wrong about “without restoring to reality any of its original dignity or status.” I think there’s a real possibility that the image world becomes SO distant from meaning that reality (and/or more intentionally curated online spaces) becomes newly enticing as the PRIMARY space to seek meaning. Very thoughtful piece! Really enjoyed reading it and will definitely return to some of these ideas.