Our threshold of repugnance

In The Civilizing Process (1939), sociologist Norbert Elias traces the emergence of the idea of “civility” to a single book by Erasmus of Rotterdam, De civilitate morum puerilium (1530), which itself underwent numerous editions and was subject to a “multitude of translations, imitations and sequels.” The book, Elias writes, “is about something very simple: the behavior of people in society — above all, bur not solely, ‘outward bodily propriety.’” It gives advice on how one should dress, how one should stand, how to eat, and how to behave in public places, going into nitty-gritty details about what to do with greasy fingers, how to deal with snot, when you should and shouldn’t fart and vomit, and so on. “Much of what he says, Elias notes delicately, “oversteps our threshold of repugnance.”

This seems like a useful way to think of TikTok. As Anna North argues in this Vox piece, TikTok has “evolved from a place to post and consume viral dances and memes into a destination for snappy, 20-second tutorials and how-tos.” North suggests that this shift is “perfectly calibrated to our particular historical moment, when many Americans are feeling insecure, adrift, and eager to shore up their self-esteem.” But it also tracks with a claim historian Elizabeth Eisenstein makes in The Printing Press as an Agent of Change (1983) about the 16th and 17th century shift to print culture: “There is simply no equivalent in scribal culture for the ‘avalanche’ of ‘how-to’ books which poured off the new presses, explaining by ‘easy steps’ just how to master diverse skills, ranging from playing a musical instrument to keeping accounts.” Think of 17th century printers, who seized and consolidated the commercial opportunities of a new means of distribution, as “printfluencers” who recognized that what the new reading public wanted most of all was to be told what to do, in any form that could be made to take.

On TikTok, then, the how-to and advice videos are not really an “evolution” from memes and participatory dances; they are all just different ways of catering to the same longstanding desire to watch and learn, to see just what it is that other people are doing, and to place that behavior into some sort of relation with our own. New media forms have always been commercialized in part through the promise that they will reveal more to us about what other people do and thereby better teach us how to live appropriate lives, At the same time as social norms are reinforced, they are experienced not as something that restricts us but as something we can enjoy as content, packaged in some new way that highlights their value as entertainment.

The “civilizing process” is at the same time a process of learning how to use and enjoy specific commodities. Norms are assimilated as opportunities or novelties or as the culmination of certain ways of paying attention that have been steadily brought into focus. Consider Elias’s description of the development of cutlery:

Nothing in table manners is self-evident or the product, as it were, of a “natural" feeling of delicacy. The spoon, fork and napkin were not invented one day by a single individual as technical implements with obvious purposes and clear directions for use. Over centuries, in direct social intercourse and use, their functions became gradually defined, their forms sought and consolidated. Each custom in the changing ritual, however minute, was established infinitely slowly, even forms of behavior that seem to us quite elementary or simply “rational,” such as the custom of taking liquid only with the spoon.

It seems equally difficult to imagine that people wouldn’t automatically know how to enjoy entertainment — that is “elementary” how to imagine the things they read about or extract pleasure from looking at pictures. But as with spoons, such self-evidence in how to enjoy those commodities is the product of a long period of social development that feels natural only when it culminates with ubiquitous objects. Knowing how to use a spoon is a basic indicator of belonging, marking that you have inherited the centuries of social intercourse and have been included in society’s ritual practices; this makes using a spoon feel right and good, and not some strange imposition being forced upon us dictatorially by “society.” Similarly what TikTok or any of the other media platforms is for seems self-evident only after their popularity can no longer be challenged, and it becomes unthinkable, for instance, that they can simply be banned by government fiat. They are not “for” any specific thing any more than table manners have some definitive utilitarian purpose; they are “for” the process of their own elaboration.

We have learned to understand what entertainment products are for in a similar way; they turn normativity into something that doesn’t feel strictly coercive. In the 18th century, the novel came to prominence in part by detailing how other people behaved in more or less plausible situations, but more important, they documented and afforded access to the characters’ thought processes about their situations. Readers would be exposed not merely to surface details of fashionability and etiquette but also to an entire mode of feeling (“sensibility”) that itself would become fashionable. (Richardson’s epistolary novel Pamela (1740) — sometimes described as the first literary “blockbuster” — is paradigmatic in this respect.)

This representation of feeling in words was not “natural” or self-explanatory; readers had to test themselves against it, absorbing normative ways of responding in the process. In narrating emotional responsiveness, novels taught readers what kinds of pity, empathy, and disgust were appropriate to which sorts of scenarios while also making those emotions consumable objects in their own right: Crying over a sad scene in a book proved that you had a “feeling heart” — your emotional receptiveness and responsiveness was “normal,” if not superior — but it also proved you could enjoy emotions privately as commodified experience, separate from social interaction.

In commercial entertainment ever since, learning about social behavior at a safe distance has been bound up with a specific kind of pleasure in objectifying feeling. When TikTok (like the reality television it derives from) combines voyeurism, didacticism, and escapism in a single package, allowing consumers to balance those impulses in varying ways, it extends that long developmental process, which has been steadily steered toward compulsive serial consumption. We can never learn enough about how we are supposed to act socially, even though there is more information about it available than ever. We can never feel secure and included, despite all the possibilities of connection. The pleasure that comes from feeling the “right” emotions evaporates even more quickly under the pressure of ever more mediated occasions to prove ourselves.

In the Vox article, North mentions the “sense of superiority” that TikTok viewers can experience from seeing other people shamed, and contrasts that with the “helpful” and “soothing” side of learning from the app’s instructional videos. But those two aspects are not so readily separated. Feeling superior and conforming to other people’s behavior are mixed together. The app positions viewers as bully and bullied at the same time. One can both identify with the people in videos and judge them, occupying both positions simultaneously and enjoying the specific sort of pleasure that derives from that tension. One can imagine conforming, adopting the values on display as one’s own, even as one disavows it; that is, one can oscillate between conforming on the level of the content (I’m going to try that dance) and on the level of its consumption (having sucked the pleasure from that video, I’m going to swipe to the next one). This is how TikTok can prompt a “doom spiral” driven by “feelings of uncertainty and self-loathing” without prompting people to quit using it altogether. It takes us beyond any threshold of repugnance.

It’s worth contrasting TikTok as a new medium for advice with the fantasy use cases currently being imagined for LLMs and generative AI. (You had to know I would end up here.) LLMs tend to be positioned as a kind of rejection of the line of social development that runs through printed books, novels, films, and television to reality television, social media, influencer platforms, and “content creation.” LLMs operate outside of that entire way of understanding and engaging with texts as models of a particular human intention and appropriate situational behavior. If TikTok puts a human face to every stray piece of social instruction, LLMs seem to do the opposite, offering statistically derived norms detached from specific human practice, from any particular person you could have any reason to identify with or judge. The models impose their normativity without any compensatory feeling of mastery or belonging, no “struggle for recognition” whatsoever.

I think this is why everything I have ever seen LLMs produce feels so empty and unconvincing. ChatGPT seems like a spoon with a hole in it. The Verge just published this comparison of usefulness of the various chatbot search engines, but it seems to skip past the question of whether it’s conceptually possible that they can be useful at all. That’s not just because they unpredictably concatenate unverifiable information, but also because they structure communcative action as a void, as outside Elias’s civilizing process. While reading the Verge piece, I could only wonder why anyone would prefer any chatbot to ordinary search, let alone the opportunity to search TikTok and see some person invest themselves in a particular piece of information.

In the article, Jay Peters describes asking ChatGPT for a marathon-training regimen and finds that the best it can do is link to an article written by a person. “If I had just wanted a chatbot to tell me what to do,” he writes, “I would have been disappointed.” But who would ever want that? It’s easy to imagine wanting to watch TikToks made by runners for all sorts of reasons (how did the training regimen make them feel? do they seem trustworthy, or like a fitness zealot? what do they wear? can I imagine myself fitting in with them?), but who wants a chatbot to synthesize some banalities about it? My imagination is limited, but I couldn’t really come up with a single significant scenario with any stakes where I would choose the chatbot over a person telling me what to do. (Of course, ChatGPT would necessarily generate some suggestions.)

Having a chatbot tell you to do anything can only be disappointing because the chatbot doesn’t care what you do. As ideas are abstracted from actual people in specific contexts, they become necessarily generic and functionally meaningless; there is no belief behind it, no practice, and this is often what we are looking for: situated knowledge, social conduct, how these are intertwined.

Elias thought that crossing the threshold of repugnance was necessary. “In considering this process of civilization, we cannot avoid arousing feelings of discomfort and embarrassment,” Elias writes. “It is valuable to be aware of them.” It allows us to see the civilizing process as ongoing, something that is possibly without an end goal or an inherent purpose.

The "civilization" which we are accustomed to regard as a possession that comes to us apparently ready-made, without out asking how we actually came to possess it, is a process or part of a process in which we are ourselves involved. Every particular characteristic that we attribute to it — machinery, scientific discovery, forms of the state or whatever else — bears witness to a particular structure of human relations, to a particular social structure, and to the corresponding forms of behavior.

My repugnance for generative AI seems related to this. The companies promulgating it offer it as a kind of ready-made “civilization” of a sort, an ultimate “general” source of the collected wisdom of the species. It extracts the minuscule infinitesimal ways that Elias posits as the ways we are “part of the process” and suggests that our activities can now be made superfluous; they can be simulated, predicted. We can consume “civility” without undertaking it.

But what bears witness to the current “structure of human relations” is not what the models generate but the very fact of their existence, a “scientific discovery” that testifies to a long developing process that was allowed to continue because it repeatedly appeared as something else, as incremental improvements in efficiency and convenience, as inevitable technological advances. That societal acquiescence seems like the relevant and telling “form of behavior” to consider, not any of the models’ simulations — that is the process we have been a part of. The discomfort and embarrassment I now feel when I encounter these models is my attempt to disavow my complicity with the “social structure” and the “forms of behavior” that put us in a position to have to deal with the power of tech companies, and the complacency with which I enjoyed how the process wore me down.