Nothing, Forever: Remix Aesthetics for the AI Age
Jake Pitre
Jun
2024
See Figure 1
In 2014, the filmmaker and editor LJ Frezza created a short video called "Nothing," which purports to contain "every shot from "Seinfeld" (1989-1998) where nothing happens." In other words, it is a supercut remix of the series consisting entirely of establishing shots or other brief moments that have no people present. As Frezza explains in the video description, "A response to everyone I met in New York who said they'd love the city if only there weren't so many people in it" [1].
As many noted upon the video's release, it takes on an eerie, almost sinister register as it continues, pushing against our expectations of presence and absence in cinema through the internet age's form of the supercut. A supercut, of course, is a collection of filmic fragments, spliced together into something new, making meaning through a hyper-intensified remix of Eisensteinian montage, updated for the attention economy. Nowadays, it can take the form of a fancam, which serves as a way to rapidly show your appreciation for a particular actor or character, or it can be more like a video essay, or it can be something more experimental.
As Max Tohline chronicles in his own video essay, “A Supercut of Supercuts,” published by Open Screens, this goes back to the first decades of cinema, particularly in newsreels, agit-prop documentaries, and other experiments with montage. As he explains, the supercut “embodies database-thinking,” whereby “the supercut was never just a subset of remix, compilation, or fan culture, but was instead a material expression of a new historical mode of knowledge and power: the database episteme” [2]. The supercut as an aesthetic form, then, is a simulation of computer search operations, and it serves the human desire for patterned repetition.
This form reached its cinematic zenith, we might say, in Damien Chazelle’s Babylon (2022), which ends with a supercut of film history alongside flourishes of film stock and its bleeding colours. The remix logic here is rather straightforward, repurposing famous images to create a new meaning of sorts. In this case, as in countless Academy Awards montages over the years, to emphasize the diversity and range of filmic expression and associated meanings, a reverence for the medium itself.
More recently, following the popular emergence of generative artificial intelligence tools, we have seen far stranger instances of database-thinking. For instance, there is the AI livestream experiment of "Nothing, Forever," (Figure 1) which features low-res Seinfeld characters spouting AI-generated dialogue until, at one point, it naturally devolves into total nonsense, including transphobic gobbledygook. “Nothing, Forever” is particularly representative of the new AI era, as it makes use of ChatGPT, Stable Diffusion, DALL-E, and other generative AI tools, an appropriate mashup and assemblage of inputs resulting in a highly fragmented output that is additionally impacted by fans interacting with the livestream via chat and various untrackable biases buried somewhere deep in the AI code.
There are complicated and fascinating relations between these two remixed products, both of which make use of Seinfeld as a repository of data, either of images taken straight from the series or of dialogue and signifiers fed into various AI servers to “create” a brief viral sensation.
As Tohline goes on to argue, in a time of endless data, “the supercut entails not simply a
mode of editing, but a mode of thinking expressed by a mode of editing” [3]. It is an ideological framing for how we experience remix in its different forms, and he suggests that our predilection for database-thinking has been a part of cinema’s representational politics since the beginning.
This sets the stage for how I tend to approach the ongoing debates over generative AI and art, because I’m not particularly interested in the question of, say, “can machines be creative?” A better, more constructive question might be what Shane Denson (2020) points to in his articulation of discorrelation and our “transition into a world of media not cut to human measure” [4]— is that the future we want? Is it the one we are being forced to accept, to adapt to, to get used to?
A simple place to begin is to point to the authorial intent of Frezza’s video, as it is a collection of images edited and remixed together, as the accompanying soundtrack eventually drifts into silence. This welcomes interpretive work on our part, because we are confronted by an object that is also a cipher for what we bring to it. The emptiness that it emphasizes is at once a commentary, perhaps, on the structures of network television, but also about this particular show“about nothing.” We can apply aspects of our own nostalgia to it, we can think through the actual construction of the short, we can read into it endlessly.
On the other hand, "Nothing, Forever" is suggestive of a future for media that transcends this existential register expressed through the remix, and brings us wholly into database-thinking and discorrelation, a truly nihilistic project that illustrates what the artist Hito Steyerl has called “mean images” [5]. You may be familiar with the writer Ted Chiang’s (2023) canny description of ChatGPT as a “blurry JPEG of the web” [6], but for Steyerl, we need to remember that these outputs are always statistical renderings, not images of actually existing objects, which means they are nothing more than probabilities.
As she puts it, “these renderings represent averaged versions of mass online booty[…]They converge around the average, the median; hallucinated mediocrity. They represent the norm by signaling the mean. They replace likenesses with likelinesses. They may be ‘poor images’ in terms of resolution, but in style and substance they are: mean images” [7]. She uses mean images as a term quite deliberately, as it signals how these images represent the mean, the average, but they are also just plain rude.
To illustrate this, she punched in “an image of hito steyerl” into Stable Diffusion, and the result was unseemly (Figure 2). As she says, “It looks rather mean, or even demeaning; but this is precisely the point…This is an approximation of how society, through a filter of average internet garbage, sees me. All it takes is to remove the noise of reality from my photos and extract the social signal instead; the result is a ‘mean image’, a rendition of correlated averages—or: different shades of mean.” But, importantly, we shouldn’t think of what this AI does as random, per se. Instead, “They are predictable products of data populism” [8]. This is database-thinking within aesthetic confines, inevitably tied up within the exploitative and harmful technical infrastructures that sustain it.
“Nothing, Forever” still streams 24/7 on Twitch since debuting in December 2022, but its trajectory now somewhat resembles that of Frezza’s short, as the more recognizable joke structures at the beginning have generally devolved into mostly silent characters making nonsensical movements over and over and over again — really what Steyerl referred to as “hallucinated mediocrity” [9]. It feels like an AI ghost town.
This should not surprise us, as it very crudely demonstrates the ouroboros that generative AI models represent, feeding on the same datasets and regurgitating itself on a self-referential endless non-linear loop of eternal soulless destruction and rebirth — what is called model collapse. These are mean images just as Steyerl describes, correlated averages that eat themselves, an abyssal abstraction.
As theorist and artist Joanna Zylinska summarized in her 2020 book AI Art:
Kindly put, much of generative AI art celebrates the technological novelty of computer vision, fast processing power and connection-making algorithms by regaling us with a dazzling spectacle of colours and contrasts as well as the sheer volume of data. Unkindly put, it becomes a glorified version of Candy Crush that seductively maims our bodies and brains into submission and acquiescence.
She adds, “It really is art as spectacle” [10].
Without going deep into Debord or Deleuze, suffice it to say that there is a long history within cultural studies and beyond about considering the machine as a thinking apparatus and as one that is capable of, if not expressly designed for, spectacle. Moreover, photography itself and the act of image-making has long been theorized as being an act of mixed authorship or co-creation, between photographer and camera, human and machine.
If we hold fast to our relational understanding of remix aesthetics and the power of images, in that very earnestly reverential manner that we’re so used to, the apparent ascendance of these AI-generated “mean images” suggests an aesthetic impoverishment that enriches those who own the digital infrastructure we use. This scenario seems unsustainable both in cultural terms, sure, but also in material, physical, environmental terms, and this intertwinement is what gives me some measure of hope in the future of media.
In his conclusion, Tohline imagines “decoupling database-thinking from its service to capitalist systems of control” and argues “that now is the time to imagine what this liberation of the database episteme could look like, both in the supercut and across culture more broadly” [11].
Likewise, Steyerl asks:
Why not shift the perspective to another future—a period of resilient small tech using minimum viable configurations, powered by renewable energy, which does not require theft, exploitation and monopoly regimes over digital means of production? This would mean untraining our selves from an idea of the future dominated by some kind of digital-oligarch pyramid scheme, run on the labour of hidden microworkers, in which causal effect is replaced by rigged correlations. [12]
Perhaps, then, it’s as simple as reversing the lingering spectre of the death of the author which continues to haunt us, inviting the return of the interpretive power inhabiting causal effect as an aesthetic constitution, a reminder that the image as such is not, or cannot be, merely a product of data populism, but one of expressive imagination. Even a humachinic act of co-creation must operate beyond the rote mechanism of the prompt. Anything else is an imposition.
Whether we think of it as rigged correlations or discorrelation or otherwise, generative AI models seek to save us the work of imagination, and their images are not only dully one-dimensional but they are an inference of an inference. The question confronting us remains how we can liberate database-thinking from capitalist control, how we can bring about that alternative future that reappropriates the means of artistic production and untrain ourselves to leave behind the philosophy of “nothing, forever” and fight back to preserve something, now.
Footnotes
[1] LJ Frezza, “Nothing,” (2014), https://vimeo.com/88077122. Return to text.
[2] Max Tohline, “A Supercut of Supercuts: Aesthetics, Histories, Databases,” Open Screens 4.1 (2021), 3, https://www.openscreensjournal.com/article/id/6946/. Return to text.
[3] Max Tohline, “A Supercut of Supercuts: Aesthetics, Histories, Databases,” Open Screens 4.1 (2021), 3, https://www.openscreensjournal.com/article/id/6946/. Return to text.
[4] Shane Denson, Discorrelated Images (Durham, NC: Duke University Press, 2020), 3. Return to text.
[5] Hito Steyerl. “Mean Images,” New Left Review 140/141 (May/June 2023), https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images. Return to text.
[6] Ted Chiang. “ChatGPT is a Blurry JPEG of the Web,” The New Yorker, Feb. 9 2023, https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web. Return to text.
[7] Hito Steyerl. “Mean Images,” New Left Review 140/141 (May/June 2023), https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images. Return to text.
[8] Hito Steyerl. “Mean Images,” New Left Review 140/141 (May/June 2023), https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images. Return to text.
[9] Hito Steyerl. “Mean Images,” New Left Review 140/141 (May/June 2023), https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images. Return to text.
[10] Joanna Zylinska, AI Art: Machine Visions and Warped Dreams, (Open Humanities Press, 2020), 76. Return to text.
[11] Max Tohline, “A Supercut of Supercuts: Aesthetics, Histories, Databases,” Open Screens 4.1 (2021), 3-4, https://www.openscreensjournal.com/article/id/6946/. Return to text.
[12] Hito Steyerl. “Mean Images,” New Left Review 140/141 (May/June 2023), https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images. Return to text.
Jake Pitre is a PhD candidate in Film and Moving Image Studies at Concordia University. His research brings together platform studies and theories of futurity to chart the temporal technopolitics of digital capitalism.