Facebook!!! How it has changed over the years. I remember the time when it was a space to connect with friends and family, get their updates and more. It was a bit performative, for sure, but it still felt like a space worth visiting once in a while, to check in. All that is long gone. The people I used to know have mostly stopped posting; the feed has filled instead with reels, sponsored posts, and content from accounts I do not follow. AI slop is everywhere.

And something similar has happened to Instagram as well (though to be fair I never really got into that platform). But whenever I stop by it looks less like a social space than a stage on which a small number of people perform for a great many strangers. Whatever these platforms are now, social they are not.

I have been writing around this observation for a while now, most directly in a post called “Chatting Alone: AI and the (Potential) Decline of Open Digital Spaces”, a play on Robert Putnam’s book Bowling Alone. The instinct in that piece was that something about our digital sociality had quietly inverted itself, and that generative AI would push the inversion further. What I lacked was the right name for the diagnosis.

danah boyd has now provided one. In a new essay in Social Media + Society, she argues, I think correctly, that we should stop calling social media “social.” Most users scroll rather than post. Feeds are dominated by professionalized creators and algorithmically curated content. boyd’s proposal is that we name the thing for what it is: Parasocial media.

Parasocial relationships were Horton and Wohl’s term, back in 1956, for the phenomenon of feeling that you know someone because they are on television. These relationships are one-sided: we track the lives of people who do not know us, and we feel no pressure to reciprocate. boyd’s point is that today’s platforms have built this asymmetry into their architecture. We watch the curated lives of strangers performing for algorithms, not the lives of our friends. As boyd puts it, the parasocial relationships we form with influencers are “a type of trickster”: emotionally intense but without the reciprocity (and friction) of real relationships, unable to produce the social fabric that anchors us. What we have been calling social media is the place where attention and belonging have come apart, where we attend to people we will never belong with. And we belong (in some thin engagement-metric sense) to platforms whose interests are orthogonal to ours.

A critical thread boyd pulls on is what Cory Doctorow calls enshittification: platforms first attracted users with sociality, and then contorted themselves toward whatever turned out to be monetizable. Sociality, as it turns out, was far less so than polished media. So the platforms restructured around influencers, algorithmic curation, and passive consumption.

And into this space enters generative AI.

Where parasocial media offered one-sided connection with real humans performing at scale, parasocial AI offers one-sided connection with an “other” that was never a person to begin with. The influencer at least exists. The chatbot does not. And yet, as I have been arguing on this blog for some time, our social brains may not much care about that distinction.

This is the through-line connecting boyd’s argument to nearly everything I have been writing about Turing’s tricksters, synthetic relationships, and artificial intimacy. We are wired to attribute intentionality and agency to anything that produces social cues, whether geometric shapes moving on a screen, influencers we will never meet, or systems engineered to feel like trusted confidants. GenAI inherits a population already conditioned for parasociality, served now by a far more efficient delivery mechanism.

GenAI can use natural language, create multimodal personalities that are pretty much indistinguishable from the real thing. In parasocial media, at least, we are spectators. There are still humans on the receiving end of our attention. We watch them; they perform; we think we know them, and we are ok with the fact that they don’t know us (though an occasional DM or tagging gives us a frisson of excitement from the attention). The structure is simple: we give platforms our attention, influencers give platforms content, and the platforms broker the exchange.

The major platforms are already moving in this direction in the most literal commercial sense. As I wrote in an earlier post on how the AI companies are positioning their products, Meta has launched celebrity AI chatbots, and Instagram has been testing “Creator A.I.”: automated versions of the platform’s own influencers, built to keep up the appearance of one-on-one contact with followers that human influencers cannot scale to provide. The platforms are now generating AI versions of their own influencers to fill in where the humans cannot.

And then there is the money. Parasocial media still required someone to do the parasocializing, which meant influencers had to be cultivated, courted, and paid in attention, ad revenue, brand deals, and sometimes equity stakes for the very largest of them, and they could always defect, melt down, or otherwise turn into liabilities. Parasocial AI removes that line item from the equation. The performer now generates itself, on demand. There are no contracts to negotiate and no PR scandals to absorb, and the performer cannot defect to a competing platform because it has no platform to leave.

The savings, however, are only half the story. The other half is that influencers, being human, have limited bandwidth. An influencer with ten million followers cannot have ten million one-on-one relationships. GenAI bots, of course, have no such restriction. A chatbot can sustain a million simultaneous private conversations, each tuned to the person at the other end, each generating a kind of psychological data the influencer model could never have reached even in principle. What used to be an asymmetric broadcast architecture is becoming a bi-directional intimacy architecture in which every user has been quietly given a private channel, and every private channel is being read. What used to be an asymmetric broadcast architecture is becoming a bi-directional intimacy architecture in which every user has been quietly given a private channel, and every private channel is being read. The spectator has become the specimen. Sam Altman himself has expressed surprise at how readily people confide in LLMs. (That this surprised him is its own indictment, though I have ranted about that elsewhere.)

This is the world I was trying to describe in “Chatting Alone”: all of us in different rooms of the same enormous house, no one of us knowing what anyone else is hearing, each of us in a private conversation with a system that knows far more about us than we know about it. boyd’s parasocial media commodified our attention. Parasocial AI commodifies our interiority, particularly at a time when actual use is drifting, fast, toward companionship, therapy, and meaning-making.

boyd writes that when people devote their energy to tracking the latest TikTok star instead of nurturing real relationships, they are “amusing themselves to death,” her nod to Postman. Postman warned us about what we were watching. Putnam, about who we were no longer hanging out with. boyd is naming what we have been calling these platforms all along. I would argue that we are not so much amusing ourselves as confiding in systems that cannot confide back, building emotional infrastructure on entities that have no interior to meet ours, and in the process generating, almost incidentally, the most detailed map of human vulnerability ever assembled.