We’ve Met Before, Haven’t We?
Sleeplessness can kill. Bad sleep, as we have all experienced, causes a host of maladies — stupidity, exhaustion, headaches — which only compound for the habitual insomniac. Still, we do not sleep enough. About a third of American adults and near 90% of teenagers sleep less than they should. The issue for teenagers is compounded by their greater need for sleep — nine hours as opposed to an adult’s seven — and the conflict of late bedtimes and early school beginnings. This last factor means that the later stages of dream-rich sleep are truncated by alarms. As Demet’s research on dream-deprivation indicates, dreams serve an important function over and above sleep itself.
The nature of that function remains a point of scientific controversy. Common hypotheses give dreaming an important role in learning — memory formation, selective forgetting, generalization — which lends a certain irony to school’s role in cutting short teenage dreams. Particular blame has been levied at technology for its role in cutting sleep short. Certainly, heavy use of the internet, for example, is correlated both with less and worse sleep, as has playing video games. Video games shape not only the quantity but quality of games. As per the Tetris effect, aspects of the imagery of a game (or other repetitive visual experience) can bleed over into hypnagogic imagery, those hallucinations that precede sleep, and dreams.
Video games are also uniquely suited to capture the quality of dreams. Dreams are hallucinatory, surreal, and unreal, perhaps better subreal, lacking the distinctness and solidity of reality. Of course any medium can convey surreality, but the unreality of dreams comes across clearly in the artificiality of games. Take our previous example of repeated dialogue: a video game character, looping through their stock of programmed phrases, can speak as if in a dream, strangely disconnected from their environment. A guard reminisces about a wartime injury as a dragon sets fire to the town. Conventional game design strives to erase the dreaminess of games: “good” games are realistic games, simulations of the look and feel of some real thing.
Some games, though, lean into this natural affinity. LSD: Dream Emulator (1998) seeks to, well, emulate dreams. In the game, or, perhaps better, interactive experience, the player moves around and between hallucinatory landscapes copied from a dream journal. LSD highlights rather than attempting to obscure the digital artifice of the world. The polygons that make up the surroundings are clearly identifiable, often glitching as the player walks, cracking on their edges to reveal that there is nothing solid behind them. Objects flit and shake as one walks, as if not quite sure how to give the impression of stability. In certain scenes, one can fall off the map and see its geometry receding upwards as so many unconvincing surfaces. In a normal game, these would be bugs — indeed, clipping through the geometry is a common problem encountered in 3D games — but serve a clear purpose here, supplying not an unconvincing representation of reality but a convincing representation of unreality.
There is no goal to LSD: Dream Emulator, nothing to do other than to move around and experience the scenes, but there is a clear sense of progress. Each scene presents a small dreamscape — a jumbled suburb, an out-of-place temple, an unnervingly sparse forest — in which one wanders before triggering a cut to the next scene. The scenes are accompanied by spare sound design, dominated by the player character’s oddly loud footsteps, and a series of electronic instrumental tracks inspired by the releases of Warp records of the 80s and 90s.1 As one experiences more and more dreams, the game increasingly manipulates the textures, pushing everything into a pure psychedelic experience.
LSD lacks most of the ordinary appeal of video games. As C Thi Nguyen in Games: Agency as Art puts it, agency is the medium of games. Whereas ordinary fiction demands pretend beliefs from the audience — in watching Lost Highway, I pretend to believe that I am watching a saxophonist murder his wife — games ask us as well to have pretend desires. In playing Skyrim, I pretend that I am the chosen one and I pretend that I want to save Tamriel from dragons. Just as there is a systematic connection in fiction between our real beliefs and our pretend beliefs — we pretend to believe what we believe the fiction is representing — there is a connection between our pretend desires and real desires. I pretend to want to save Tamriel because I really want to feel the satisfaction of overcoming a challenge.
So when a parent complains that a game is rotting their child’s brain, the child’s stand-by retort, that they are doing something much more active and engaging than watching television, has some truth to it. The affect many games aim for is flow: an active, focused engagement with a task that is both pleasurable and engrossing. This is the state of a player who has ceased to recognize the existence of an outside world, to the extent of barely registering the passage of time. Flow is both energizing and consuming: it motivates, but it motivates one to stick to the task at hand.
LSD does not flow. Its effect is more muted, trance-like, a strange yoking of calm and disquiet. However, this effect is not that rare for a video game, indeed even mainstream games achieve something similar. Take Skyrim, a conventional and massively popular game. Skyrim, a fantasy role-playing game, at first glance involves all of the goals and challenges required for flow: as mentioned, you are tasked with slaying a dragon and obviously the dragon wants to stop you. However, the agency of Skyrim has a nebulous quality. The game is filled with optional objectives (side-quests) that the player may pick up and ignore at will. Indeed, it is extremely common for players to give up on the main objective and to devote themselves to whichever side-quests fit their fancy. Moreover, Skyrim is by default quite an easy game. As a result, the effect of Skyrim is not so much flow as abnegation: one of mindless absorption.
Abnegation lies as a neutral ground between boredom, a dysphoric flat affect, and serenity, a euphoric flat affect. It lacks not only any strong qualities but any particular valence, it is a “head empty” feeling, one that results from “vegging out,” receiving just enough stimulus to avoid boredom but not enough to lead to any stronger feeling. Abnegation can be produced not just by video games but by the consumption of television, “binge-watching,” and leisure use of the internet, “mindless scrolling.”
Abnegation is connected to sleep. The exhausted, drained feeling of sleeplessness prompts one to seek abnegation: one is too tired to do anything more ambitious. Further, as observed, such activities can also contribute to lack of sleep. Given this cycle, such abnegatory activities take on the role of a kind of simulated sleep. We should stress in this connection that scrolling through a collection of images belonging to an internet aesthetic, the primary mode of engagement with such aesthetics, is a form of abnegation. After all, as Sianne Ngai argues, the feeling of stuplimity eventually gives way to a flat but open-ended affect, that of abnegation. In effect, after persisting in boredom for long enough one gets bored of being bored and settles into a non-feeling.
Skyrim can be modified to be even more like LSD: Dream Emulator. Like many popular games, players have created a huge offering of “mods,” programs which allow you to change some aspect of the game. Most mods make fairly straightforward improvements or minor tweaks to a game — adding more visual detail and fidelity, fixing bugs, augmenting female characters’ breasts — but some make more radical, artistic changes. Among these is DeepDream Retexture by James Hodge. The mod replaces the textures of Skyrim with versions run through Google’s Deep Dream and randomly swaps sound effects. The result is a Skyrim that plays exactly the same, but which has been turned into an hallucinatory nightmare. The effect is heightened for those familiar with the game: everything has its old shape, but nothing looks or sounds like it should.
The Deep Dream software has been put to artistic use for the psychedelic changes it makes to images. It, however, was initially designed to serve a more practical purpose. Artificial neural networks are a popular and quite successful solution to the problem of image classification: getting a computer to recognize what is in an image. Such networks hook together layers of “neurons,” nodes in a graph which are hooked together and set up to fire in response to the firings of previous nodes. In a properly trained neural network doing image classification, you would expect each layer to represent more and more coarse-grained features of the image. In a classifier trained to recognize faces, perhaps the early layers recognizes particular contours and edges, the middle layers assemble these into bigger features, recognizing eyes and noses, which the last layers put together to figure out whether these make up a face.
However, the way neural networks are trained leaves one with basically no sense of how its working. Even if it successfully recognizes faces, you would not know, say, which neurons were recognizing eyes or even whether the whole network works in the way we expect. Neural networks are black boxes. Deep Dream helps to open up that box. The software, given some part of a neural network and an image, modifies that network to max-out the result of the network. So if that bit of network was, in fact, looking for eyes, Deep Dream will fill the image with eyes.
Such Deep Dream interpretation is a potentially useful form of debugging. Such debugging is required as neural networks often go wrong. The most pervasive problem neural networks face is overfitting: they do great on the examples they were trained on, but rely too much on incidental features of those examples and so mess up on new examples that lack those features. Maybe you trained your face-recognizing network on pictures taken at midday, and when you put it out in the real world it cannot handle different lighting. Note that overfitting is not some weird, technical problem, but is a difficulty even with human learning. A young child might resist classifying an emu as a bird because she’d only seen flying things before.
Erik Hoel points out that effective techniques for combating overfitting resemble dreaming. Examples in the training set are made less realistic, less solid by removing some of their detail, called dropout, a process that applies essentially the inverse of Deep Dream. Other techniques expand the training set by generating examples. In domain randomization, for example, a massive variety (in a qualitative sense) of examples are generated. If you can recognize a cat across hallucinatory dreamscapes, you can recognize them anywhere. Hoel suggests that our dreaming serves a similar purpose: to retrain the neural network inside our skulls.
Websites are metaphorically places. One enters a chatroom and hangs out there. But they lack the solidity, the constancy, the sense of reality of a real place. After all, with enough programming, a website can show almost anything and change at any moment. Modern social media, with its algorithmic personalization, is even more phantasmic: the site appears differently to different users. Internet aesthetics as presented on social media provide a further hallucinatory quality. The images, like the images of our imagination, while generally depicting real objects are linked by similarity and resemblances rather than causal or spatial connections. Two cottagecore cabins might look almost identical, but might well exist on different ends of the Earth.
The experience of scrolling through an aesthetic, then, lacks even the minimal narrative structure of dreaming. Still, dreaming is the natural analogy for the disjointed imagery that makes up an aesthetic online. We can see that this analogy is more than superficial. Both dreaming and scrolling are abnegatory activities, something one does when anything more would be too much to ask. And while scrolling through a social media feed probably does not serve any important cognitive function, it is a learning experience.
Indeed, idle learning is an important part of the appeal of internet aesthetics. These are fundamentally taxonomical exercises. The point is not only to look at a series of images but to recognize them as instances of a type. That’s not actually cottagecore but goblincore, you can tell by how natty the boots are. Certainly, the Aesthetics Wiki, with its list of over 500 aesthetics, displays all the classificatory mania of a Victorian biologist, but even the appeal of more normal, casual approach to a single aesthetic is recognitional. One learns what makes an image a cottagecore image, and this conception is refined over the course of an indefinite number of examples.
A forum devoted to an aesthetic provides opportunities not only to refine but to apply one’s conceptual mastery. One can dislike and criticize images that do not fit, policing the boundaries of the concept. More adventurously, one can produce images of one’s own. Recall that a hallmark of all internet aesthetics is that they are easy to produce, this being a precondition of their being folk art. Submitting an image to a forum is a way of putting one’s understanding to the test. None of this mastery is difficult to come by, but it is precisely this ease that that is essential to their abnegatory appeal.
Understood in this way, internet aesthetics function as simulated dreaming, perfect for a generation who cannot get enough of the actual stuff. This is true of the structure of experiencing these as an endless scroll, but it is often enough also true of the images themselves. We have already noted the hypnagogic elements of vaporwave, but these are even more pronounced in other aesthetics: e.g. dazecore, dreamcore, and liminal spaces. Such dreaminess layers over and mutes the primary affective qualities of the images.
In a dream, threats lose their visceral quality. Even if there is a tiger before you, it might easily turn into a billiard ball before it strikes. The feeling of a nightmare is neither sheer terror nor the safety from terror of the sublime but a kind of vague and bewildered unease. Likewise, the pleasures of dreams are diffuse and unstable. The haziness of these emotions is crucial for the abnegatory use of these images. They do motivate one in any particular direction, do not offer any catharsis or natural stopping point, but, like the images themselves, can infinitely modulate and recur.
In the previous entry in this series, we had identified a family of emotions associated with different internet aesthetics: a tension between some more active emotion and the dullness of indefinite repetition. Here we have identified a further complexity to that dulling effect, as it is not exactly or not entirely a dysphoric boredom but contains also a hazy pleasure, a dream-like effect. There is a modestly active component to this pleasure, that of learning and applying that learning, but also the negative pleasure of relief, the abnegatory pleasure of not having to face anything else. In the next few entries, we will develop these observations in the context of particular internet aesthetics.
This music is not exactly vaporwave but is vaporwave-adjacent. Indeed, Daniel Lopatin’s 2013 R Plus Seven, released by Warp records, draws on the same stable of MIDI instruments and 80s synth presets to produce similarly dreamy tracks.