March 27

trompe l'oeil; social homogeneity

(I’m considering doing posts like this as part of this newsletter’s subscriber-supported expansion. Please let me know if it seems stale or redundant!)

Puffy pope trompe l’oeil

This past weekend, an AI-generated image of the pope wearing a puffy white jacket circulated online and has now been anointed the first significant Midjourney meme. It seems less a harbinger of the “end of reality” or “disinformation armageddon” than an indication that AI-generated images have become ordinary meme fodder and are now being broadly assimilated at that level. As James Vincent suggested, “it works as a fake precisely because it matches ways we already consume images today” — which, I think, is as rhetoric and not as factual documents.

When something circulates within the context of “memes,” this indicates how it is supposed to consumed: as something made deliberately to accomplish some purpose, whether that is to mock somebody or make someone laugh or to get someone riled up or whatever. It’s satisfying to be momentarily tricked by an image, satisfying too to imagine others being tricked but not you. As AI-generated images spread as low-stakes memes, they allow more and more people to grasp that the funny images that they will continue to see in social media or in message forwards that possibly trick them for a moment are almost certainly not documenting reality but displaying someone’s ingenuity in executing an idea and remixing media components with a new kind of image editor. But the fact that some new technology is being used to perform these sorts of feats doesn’t seem all that revolutionary. It’s not like AI invented storytelling.

As with previous panics about “faking” the real, much dubious nostalgia is being expressed for the pure and unadulterated access to “reality” we’ve supposedly lost with the advent of new media technologies. This assumes that reality is a matter of a representational fidelity, and that everyone in the world is stuck at some level of idiot empiricism where sense-certainty is as far as reason goes. But what is taken to be real does not simply depend on how realistic something appears; realism is itself a set of conventions that are determined contextually and socially. Facticity depends on who reports the facts and who finds that reporting legitimate; it doesn’t depend on “what really happened,” as if that were simply accessible to us with a really good piece of recording equipment.

“Reality” is a social construction that can’t be entirely short-circuited with images alone, no matter how realistic their details are. Those images would need to be taken up by the various institutions and apparatuses that work to produce consensus and validate knowledge, and even then they wouldn’t be able to reach some uncontested status beyond further revision. Or to take up the same idea from the reverse perspective, all images are “real” when people circulate them, because the communication being carried out through the images is real, because some human intentionality is being realized through that ongoing distribution. What is “real” isn’t simply empirical data and “content”; reality is a social form.

The “age of average”

In “The Psychological Structure of Fascism” (1933), George Bataille characterizes capitalist society as dependent on “social homogeneity” in which everything and everyone is the same insofar as they are construed as useful, capable of being made into something commensurate that can be used by systems of production. Anything that threatens this homogeneity Bataille describes as “heterogeneous”: these are “elements that are impossible to assimilate” and as such are almost impossible to represent but which consist of “everything resulting from unproductive expenditure” — for Bataille this tends to mainly be a matter of taboo practices, violence, and other objects or tendencies perceived as excremental, antisocial, or criminal.

That analysis came to mind after I followed a Metafilter link to an essay by Alex Murrell called “The Age of Average.” It provides a systematic overview of the homogeneity to be found across many cultural domains: “Distinctiveness has died. In every field we look at, we find that everything looks the same.” Interiors look the same; shops and restaurants look the same; apartment buildings look the same; city skylines look the same; cars look the same; brands look the same; book covers and movie posters look the same; people looks the same; faces look the same; experiences look the same.

Murrell doesn’t offer an explanation for why this is happening — usually blame is assigned to algorithmic media more rigorously imposing certain norms over wider territories and increasing the rewards of conformity — but it seems like convincing evidence that homogeneity has definitively triumphed over heterogeneity. It can also be read, however, as a sign of the increasing pressure to contain the heterogeneous, that so much time and space must be devoted to strenuous displays of sameness.

Naturally, that is how I would interpret generative AI and LLMs as well: These systems work by imposing homogeneity on everything they touch and are beginning to be integrated into everyday life to carry over that homogeneity into social practice as a kind of horizon of possibility. Sometimes this will appear as averageness and aesthetic mediocrity, but more often it will show up as ordinary productivity. But the urgency and alacrity with which companies want to impose generative models on the world shouldn’t be understood simply as a consumer demand for homogeneity, productivity, sameness, and comfortable blandness. It may indicate a sense that the homogeneous is losing its grip — that in Bataille’s terms, it is “incapable of finding in itself a motive for requiring and imposing its existence.” Hence it must be automated, as machines don’t need a motive.