Performing as Human in 2026: AI's latest twist on 200 years of Mediated Presence
An abridged history of performing for audiences you cannot see, and the new wrinkle in it. Competing against GenAI now, development of theories of audience then.
A recent Wall Street Journal piece profiled writers who have started deliberately scuffing up their own prose to prove a human wrote it. Sarah Suzuki Harvard Sarah Suzuki-Harvard, a Brooklyn copywriter, has been “going rogue” with intentionally casual phrasing and bursts of exclamation points. One writer leaves in accidental typos rather than fix them. Another replaces his em dashes with two smaller dashes that are meant to look more “handmade.”1 Andy O’Bryan Andy O'Bryan, who has written about the trend himself, calls it a reverse Turing Test.
The reverse-Turing framing2 is doing useful work. It names something. But the flattening also obscures what is actually happening underneath, which is the latest jagged step in a much longer arc, going back to the first photographic portrait, of the subject of a mediating technology learning who is on the other side of it.
That arc is worth tracing in compressed form, because the WSJ piece reads differently when you place it inside the timeline, and because the new wrinkle in it (which is real) is more specific than “humans want to seem human.”
Mediated presence: a brief history of performing to a technologically abstracted audience
Daguerreotype era, roughly 1830s through 1850s. Photographic portraiture is brand new. The medium has no prior conventions for the subject to inherit. People sit very still, partly because exposure times demand it, but also because they do not yet know what a person photographed does. The sitter’s face is uncomposed in any social sense; it is just present in front of a piece of equipment whose downstream effects are difficult to anticipate.
It is worth noticing that the early portrait subjects were not unsophisticated. They were dealing with a genuinely new genre.3 Mark Twain, who knew the medium well, refused to smile for photographs and put the reasoning plainly: “A photograph is a most important document, and there is nothing more damning to go down to posterity than a silly, foolish smile caught and fixed forever.” That is a coherent theory of mediated presence: the image will outlast you, will be seen by people you do not know, in conditions you cannot anticipate, so do not be caught looking like a fool in it. That theory is not naive. It is one of the earliest documented examples of a person reasoning about an asynchronous, dispersed, future audience.
Carte de visite era, late 1850s through the 1870s. Small, inexpensive portraits begin to circulate. People exchange them, paste them into albums, send them. The International Center of Photography has a useful framing of this period: the carte de visite was, in effect, the first widely circulating personal-image medium, and its subjects were already engaged in what Alice Marwick would much later call social surveillance. Sitters chose poses, props, and clothing with the explicit anticipation of dissemination. The photograph was no longer a one-off object for a sitting room wall. It was something that traveled. The subject had to start composing for a public they would never meet face to face.
The subject had to start composing for a public they would never meet face to face.
Kodak Brownie era, 1900 forward. The camera is now in the hands of children and amateurs. Marc Olivier’s history of the Brownie tracks how the $1 camera released in 1900 sold ten million units in five years and pulled photography out of the studio and into ordinary life. Photography stops being a destination and starts being an ambient practice. The subject of a photograph is now sometimes also the person taking the photograph an hour later. The schema for being-photographed becomes part of ordinary social competence.
Mid-century. Smiling for the camera is gradually normalized as the expected expression. Time has a clean piece on this transition: smiles do not become standard in photographic portraits until the 1920s and 1930s, substantially reinforced by Kodak’s own advertising, which trained people on what a “good” photograph of a person looked like. “Say cheese” is taught as a literacy, transmitted across generations, and runs through most of the twentieth century. The reverse-Turing test of the moment, if you wanted to describe it that way, was whether you knew how to perform appropriate liveliness on demand.
Cinéma vérité reckoning, 1960s. Documentary filmmaking has to confront, theoretically, what photographers had quietly understood for a century: the camera changes the subject. Jean Rouch and Edgar Morin’s Chronique d’un été (1961) is partly an extended argument that the camera’s presence is itself part of what the camera is recording, and that pretending otherwise produces a worse document, not a more honest one. The opposing posture, direct cinema, tries to minimize the perturbation. Both posit a self-aware subject. Both implicitly accept that the subject knows the audience exists.
The webcam and early YouTube era, late 1990s to mid-2000s. This is the period that gets flattened in retrospect, and is worth looking at slowly, because it is where the texture of audience-awareness shifts in a way that does not quite match what came before or what came after.
Jennifer Ringley’s JenniCam, which ran 24/7 from her dorm room from 1996 to 2003, is a useful baseline. Ringley was at one point being watched by millions of people, and her stated position on it was: “I don’t feel I’m giving up my privacy. Just because people can see me doesn’t mean it affects me — I’m still alone in my room, no matter what. And as long as what goes on inside my head is still private, I have all the space I need.” That is a theory of the audience that becomes almost impossible to hold ten years later. The audience is a fact, but the subject is asserting that the fact is bounded by the camera frame and does not migrate.
The first YouTube video, Jawed Karim’s “Me at the zoo”, uploaded April 23, 2005, has the same texture. It is nineteen seconds of one of the site’s co-founders standing in front of an elephant enclosure with no apparent script and no apparent reason for the audience to be there. The camera is held by someone (Yakov Lapitsky), the speaker addresses it, and the resulting artifact has none of the conventions that would become standard inside a few years. There is no opening, no sign-off, no styling, no acknowledgement that the audience could be anybody on earth. It is one person speaking to a recording device and a friend, in a way that has already become hard to do unselfconsciously.
The anthropologist Michael Wesch’s 2008 Library of Congress lecture, An Anthropological Introduction to YouTube, and the related paper the following year, took on this question directly. The paper’s title says most of what needs to be said: Experiences of Self-Awareness in the Context Collapse of the Recording Webcam. Wesch’s image for what happens when an early vlogger faces the lens is exact: “The little glass lens becomes the gateway to a black hole sucking all of time and space — virtually all possible contexts — in on itself.” His subjects are visibly struggling with a problem that did not yet have a working vocabulary, much less a developed literacy.
Home video, roughly 2003 to 2007. Watch footage of high school students from this period, the kind of clips that have surfaced on archive accounts and personal YouTube uploads. There is a particular quality of being on camera in that footage that has mostly dropped out of contemporary recording. The subjects often look, in retrospect, almost unguarded, because the schema for being filmed all the time is not yet installed. The camera was somebody’s parents’ camcorder, or the school’s news team’s, or a friend’s. Footage was watched in specific contexts. It did not yet flow at low friction across platforms and search results. The literacy of being-on-camera was situational and bounded, much closer to the carte de visite subject’s understanding of a single image traveling to known recipients than to what would come next.
Smartphones, late 2000s onward. Everyone is now a camera, always. The cultural shift is well-tracked: a recent psychological study of toddlers in what the researchers call the “selfie generation” found that 95% had been recorded on a phone or tablet in the previous month and 42% were being filmed at least weekly. That is a baseline of mediated self-presentation that previous generations did not have to integrate before they could speak. The subject’s awareness of the audience moves from situational to ambient. Being on camera is no longer an event; it is a background condition. The naïveté in 2000s high school footage is partly that the subjects in it are still reasoning about specific cameras held by specific people. Within a decade, that reasoning has to expand to anyone who might ever pick up a phone in the same room.
Platforms, 2010s onward. Erving Goffman’s 1959 The Presentation of Self in Everyday Life had already given a vocabulary for treating ordinary social life as performance with audiences, front-stage and back-stage. Danah boyd and Alice Marwick’s work on context collapse and the imagined audience names what happens when those previously separated audiences are flattened into one undifferentiated public by social media. By the late 2010s, the literacy expected of an ordinary person mediating their own self-presentation online is already wildly beyond anything a Victorian sitter or even a 2003 high schooler would have recognized.
Now. The opening Wall Street Journal piece.
From Mediated Presence to Competing for Viability and Authenticity against Generated Content
The temptation is to file this latest move as more of the same on a curve that has been bending in the same direction for almost two hundred years. Subjects become more aware of audiences; performance becomes more elaborate; literacy demands compound. By that read, the writer adding typos is just the latest iteration: an audience-aware human composing for dissemination, the way the carte de visite sitter or the early YouTuber was.
That read is partly right and partly missing the point.
What is genuinely new in the WSJ piece is not the degree of audience-awareness. The carte de visite sitter, the cinéma vérité subject, the early YouTuber, and the contemporary writer are all reasoning about asynchronous, dispersed audiences whose composition they cannot fully see. What is new is the category of audience they are now reasoning about.
For the entire arc from the daguerreotype to last decade, the audience the subject was learning to anticipate was, eventually, going to be composed of other people. Other people with eyes, other people with their own cameras, other people with cultural expectations, other people with biases, other people who might forward the photo, the video, the post. The literacy that was developing was a literacy about other human viewers, refracted through whatever technology was in between.
All the (digital) world is a stage?
The writers in the WSJ piece are reasoning about an audience that includes non-human readers and non-human writers. The pressure has shifted. Previous versions were about being seen well by an audience whose attention was scarce. The current version is about being distinguishable from an audience-adjacent population whose output is functionally infinite, whose surface conventions overlap heavily with the subject’s own, and whose presence at any given reading of a piece of writing cannot be ruled out.
The “ickiness” Sarah Suzuki Harvard described picks up on something specific. It is the experience of having to actively differentiate one’s own output from a flood of plausibly-similar output produced by an asynchronous, ambient, partly-invisible other category of writer who shares the same surface conventions a careful prose stylist would. The differentiation is not optional. The work of pre-emptively distinguishing yourself happens inside the writing itself, before the writing leaves your hands, at the level of word choice and sentence rhythm and yes, em dash placement.
That is a different kind of audience problem. The Victorian sitter worried about looking foolish in front of strangers. The contemporary writer is worrying about being mistaken for a process. The previous literacies were about composing the self for human reception. The current literacy is partly about composing the self in a way that survives non-human imitation.
It is worth being precise about which part of this is the new pressure. The imagined audience was already strange. Context collapse was already strange. The new thing is that the imagined audience now contains imagined generators: writers, painters, voices, faces, some of which can produce output indistinguishable from the subject’s own, and the subject is now responsible for differentiating themselves from those imagined generators in advance.4
The literacy required is not yet stable. It will move. The current crop of “tells” (em dashes, lists of three, certain words, certain rhythms) will not stay reliable as tells, partly because models will adapt, partly because human writers will absorb the patterns until the patterns stop being legible as artificial, partly because the detection tools are already producing false positives at rates that suggest differentiation-by-surface-style is unstable as a long-term strategy.5
We would do well to consider where the tells come from; academic and historical literature has served as training for professionals as well as AI. If hallmarks of rigor and erudition are now blurred with with inauthenticity signaling — UC Berkeley School of Law and arXiv both have taken harsh stances this month to address such — that will be an interesting problem space to tackle.
There is also a second move worth noting, opposite in direction to the typo-planting one. Ryan Johnson, one of the writers in the WSJ piece, described what happened when he tried using AI to draft his blog posts: “It’s like the restaurant that starts to water down the soup. People don’t leave immediately, but eventually they’re like, eh, it doesn’t have the same kick.” He quit. Both moves are responses to the same pressure. One adds noise on the way out to seem human. The other strips out the generative shortcut on the way in to remain human in a way the noise-adding move is already conceding cannot be done from the surface alone.
Two things are likely to follow.
The first is a continued, escalating round of surface-level mimicry-and-counter-mimicry. The “say prunes” of 2026 is the deliberate typo, the planted Office reference, the artisanal short dash, the unstructured aside that breaks the rhythm a model would produce. Some of it will work for a while. Most of it will get absorbed into the model distribution within a year or two. The signal-to-noise of stylistic tells has already started to invert, and there is no reason to think the curve flattens any time soon.
The second is more interesting, and more durable. The non-surface literacies for being human at this moment will become load-bearing in ways they were not before. The capacities that are hard to fake at scale (sustained argument that builds across a body of work, claims grounded in specific operational experience, intellectual positions held across years and revised in the open, the kind of through-line that takes a person to develop and a person to keep alive) become more valuable, partly because they are increasingly the things that can still differentiate human from synthetic at altitude. Johnson’s instinct, that watered-down soup gets noticed eventually, is the version of this argument that does not require theory.
This is the part that connects to the broader project I have been writing about. The arrangements we inherited were legible because they had stabilized into recognizable forms. The arrangements we are entering are not yet legible, and the work of the next few years for anyone composing themselves into a mediated, asynchronous, partly-non-human audience is partly the work of figuring out which literacies will hold and which will be obsolete inside eighteen months. Furthermore, how do you meaningfully communicate with others, as the means to do so and barriers remain in flux?
Addendum on this year’s flash-trend of students booing graduation speakers who are celebrating AI
This pressure is not staying contained inside the small world of writers and prose stylists. The class of 2026 has been booing commencement speakers every time they mention AI. Gloria Caulfield got booed at the University of Central Florida for calling AI “the next industrial revolution.” Scott Borchetta got booed at Middle Tennessee State for telling graduates, “It’s a tool. Make it work for you.” Eric Schmidt got it the worst, drawing sustained boos at the University of Arizona to the point where he paused to say “I can hear you” mid-speech. Glendale Community College’s president drew boos just for explaining that an AI system was reading graduates’ names from the stage, partly because the system had been getting the names wrong.
Some of this is generational6 defiance, and that is part of it. Underneath, though, the graduates are reasoning about the same pressure the writers are, from the opposite end of the production relation. The writer adding typos is trying to stay legible as a human inside an audience that now contains non-human producers. The graduate being told to welcome the category of producer currently competing with them for the meaningfulness of their own labor is registering that this is not a welcome at all. Mistaken for the system in one case; replaced by it in the other. Same underlying shift in who and what the audience now contains.
Mark Twain’s worry about being damned to posterity by a foolish smile is, in this light, an old version of the worry Sarah Suzuki Harvard names when she calls the contemporary version “icky.” Both are reasoning, in the available vocabulary of their moment, about an audience whose composition they can only partly see, and about how to compose oneself for that audience without losing whatever it is about being a person the audience is supposed to be receiving.
-
Closing “Two-Tracks” Thought: It is worth wondering how different communities will emerge, with both conventions and infrastructures to center the meaning and conversations they aspire to see more of in the world.
Appendix: A short reading list for the long arc
For anyone wanting to follow the thread further, these cluster into a few literatures that rarely get put in the same room, which is part of why the arc is hard to see at once.
On the photographic subject and the gaze. Susan Sontag’s On Photography (1977) is the foundational treatment of the photograph as object, the photographer’s relationship to subject, and the strange new relation people now have to their own image across time. John Berger’s Ways of Seeing (1972) and About Looking (1980), the latter containing “Uses of Photography,” are written partly in response to Sontag. Roland Barthes’ Camera Lucida (1980) is on the photograph and the looked-at self; the source of the punctum concept. Tamara Berghmans and Ingrid Leonard’s Early Gaze: Unseen Photography of the 19th Century (FOMU, Antwerp, 2025) is a recent curatorial treatment of the awareness arc in early photography.
On performance, audience, and self-presentation. Goffman’s Presentation of Self (linked above) predates social media by decades but reads now as eerily prescient. Marwick and boyd’s 2011 paper (linked above) is the canonical work on context collapse; boyd’s blog post on the term’s origins is useful supplementary context. Michael Wesch’s 2008 lecture and 2009 paper are the cleanest contemporaneous account of the early-webcam audience problem. Sherry Turkle’s Life on the Screen (1995) and Alone Together (2011) trace the longer history of identity-formation through screens.
Additionally, Amanda Nicole Curtis Amanda Nicole Curtis raised many insightful points when drafting this piece, including mentioning these names for the Appendix:
Haidy Geismar’s work offers us a theory of ‘object lessons’ building on work by Lorraine Daston: ‘Object lessons are… both ontological (they tell us something about what there is) and epistemological (they help us interpret and explain what there is)’. Geismar reminds us that: ‘As much as digital media brings new ways of looking at and understanding collections, it also represents, and refracts, earlier representational techniques’
On the camera as theoretical problem. Chronique d’un été (Rouch and Morin, 1961) is the film itself as argument; cinéma vérité as a posture lives or dies on whether the camera’s presence is honest material rather than perturbation to be hidden. Stella Bruzzi’s New Documentary (2nd ed., 2006) tracks the subject-aware-of-camera question through the late twentieth century.
On the democratization of the camera. Marc Olivier’s “George Eastman’s Modern Stone-Age Family: Snapshot Photography and the Brownie” (Technology and Culture 48(1): 1–19, 2007) on the 1900 Brownie and the move of photography into ordinary hands.
On the webcam and early YouTube period. The POV Magazine retrospective on JenniCam, Petra Cortright, Natalie Bookchin, and others is a useful entry point. An Exhaustive (But Incomplete) History of the Webcam on Highsnobiety is broader-strokes but covers the cultural arc well.
On the WSJ piece itself and the present moment. Te-Ping Chen’s original Wall Street Journal piece. Andy O’Bryan’s response. O’Bryan’s framing of the moment as existential as well as stylistic is the right reading of what is happening underneath the typo-planting.
The arc is not a single literature, and no one of these texts captures the whole thing. The most interesting work is going to happen in the room where they are read together, which is mostly not yet a room that exists.
But there are many working to build that room, and the associated literacies. If you know folks or recommend any related resources, please leave a comment below or reach out.
A handful of contexts. Also of note: latency and perceived thoughtfulness, [2]; an arXiv paper on hesitation and self-editing in chatbots; and commentary on how Google’s NotebookLM was made, pauses and all; not generated by the language model in the transcript — they're deliberately built into the audio model itself (Latent.Space). Bonus: an associated lawsuit on AI Voices.
A general history-of-science public service is a reminder to contextualize Turing’s actual lived experience “passing.” I wonder what Turing himself would say about human rights, “AI rights”, and the current pressure to find ways to not pass as AI.
The fullest popular framing of the Victorian non-smile is Olivia B. Waxman, “Why People Didn’t Smile In Old Photographs,” Time — the smile-as-standard-expression transition happened in the 1920s and 1930s, much later than most assume, and was substantially shaped by Kodak’s marketing. The dental-hygiene story is real but is doing less work than it gets credit for; the social etiquette around smiling, and the inherited convention from painted portraiture that grinning was for fools and drunks, is doing more.
This is also rapidly expanding issue in various submissions, applications, and even pull requests. In a recent meeting of the Google Summer of Code + OpenSource Summer Cohort, Orthogonal Research and Education Lab’s Dr. Bradly Alicea Bradly Alicea led a discussion on the slop deluge experienced by many organizations this year.
The false-positive problem is well-documented: a 2023 study found major AI detectors flagged non-native English writing as AI-generated at rates far above native writing, and the WSJ piece touches on how the standard of "human-written" is itself drifting toward the model distribution as people unconsciously absorb its patterns. Ivan Jackson, the Writehuman founder quoted in the piece, notes that human-rewritten text is increasingly being flagged as AI. That is the leading edge of the differentiation problem becoming circular, and it is one of the reasons surface-style tells are not a stable long-term differentiator. (Also, I see Grammerly is offering a “AI Humanizer” tool now, as well.)
It may also be worth noting, for historical and contextual purposes, the sentiments captured in this piece by Chronicle of Higher Education: “Where are all the student protests?” There is a particularly memorable indignity for (American) students who have been so policed in the last few years — now, met with a nudge to stand up and have a smile on your face about AI, particularly jobs [2], as well. The inability to comprehend this aftermath is telling, in a way. It is a strange position to be in, as someone at the nexus between: young folks looking for meaningful things to do (or just get paid); seeing businesses contend with AI implementation pressures; and seeing the gaps that are relatively unattended by older generations. In the Sagan-foreboding quote sense, the ‘generationally underprepared’ are looking for things to do, and have increasingly less incentive to carry water for the systems and structures that produced the current mockery — let’s say, difficulty — of their circumstances.
An insightful comment from Darren Lott of the Brain Inspired Complexity Discussion Group: "Performing for audiences you cannot see” captures much of the terror of public speaking. With a small present audience you may be able to judge how happy they are with your performance and adjust accordingly. It becomes more out of control as the audience size grows.
But even more disconcerting is performing for audiences that don’t currently even exist. This may be appearing in a stylish manner that a decade later is laughably out of fashion, or making a comment online that decades later, under a new social sensibility, prevents you for getting or keeping your job."