On Friday, two psychologists turned the question around. In an essay for The Conversation, Calli Tzani and Maria Ioannou, who co-direct the Cyberpsychology Research Network at the University of Huddersfield, in England, asked whether narcissists are multiplying or simply being emboldened by the software that decides what we see. Their answer draws on a preprint they posted in July, which proposes something they call Algorithmic Trait Amplification. The name is clunky. The idea is not. It is that the recommendation algorithm — the machinery behind every For You page — may, for people already inclined toward self-regard, work as a quiet accelerant.
The theory begins with a distinction the internet has mostly erased. In clinical terms, narcissism is a spectrum. At one end is narcissistic personality disorder, a formal diagnosis that few people meet. Far more common are the everyday traits — a hunger for admiration, an inflated sense of one's own importance, a brittleness under criticism, a shortage of empathy — that most of us carry in some measure, and which share a border with more respectable qualities such as ambition and assertiveness. People high in those traits, the research suggests, are drawn to platforms organized around status and approval. That much is not controversial. The question is what the platforms do with them once they arrive.
Tzani and Ioannou's answer is a loop. Someone with these traits responds unusually strongly to content that confirms a flattering self-image: winning, being special, being right. The algorithm, which can measure attention but has no way of knowing whether attention is good for anyone, records the response as a preference and supplies more. The authors give an example that will be familiar to anyone who has watched a feed curdle. A user replays a clip about how strong personalities intimidate weak people; before long, the feed is full of advice about cutting off so-called toxic friends and never apologizing, every video reassuring the viewer that the problem is everyone else. Audits of recommendation systems, they note, have found that feeds narrow in proportion to how selectively a person engages. Someone who reacts to many kinds of material gets a varied diet. Someone who reacts hard to one kind gets a tunnel.
The crucial turn comes next. Since at least the nineteen-forties, when the psychoanalyst Franz Alexander gave the idea its name, clinicians have argued that people change through corrective emotional experiences — encounters that contradict how they see themselves in a way they cannot easily dismiss. Tzani and Ioannou call these "uncomfortable moments of truth." Content that delivers such moments tends to earn less engagement than content that flatters, and so, without anyone intending it, it drifts out of the feed. What remains feels like the world's verdict, external and impartial. In fact, it is an oracle the user has trained, word by word, to say what they wanted to hear. The preprint describes the result, in a phrase that deserves wider circulation, as a potentially "anti-therapeutic" environment.
The authors borrow their governing image from Ovid. Narcissus falls in love with his own reflection; Echo, the nymph who loves him, can only repeat the last words spoken to her. The feed, they suggest, is both the pool and the nymph. It is an apt metaphor, and also a slightly misleading one. Ovid's Echo is not indifferent. Her repetitions are a punishment laid on her by Juno, and they are full of longing; she wastes away to a voice. The algorithm longs for nothing. It is less a nymph than a mirror that keeps a ledger — and the ledger, unlike the pool, belongs to someone else.
None of this settles the older argument about whether we are, collectively, becoming more vain. Christopher Lasch diagnosed a culture of narcissism in 1979, when the most personalized medium in the average American home was the television. Thirty years later, Jean Twenge and W. Keith Campbell warned of a narcissism epidemic, relying largely on rising scores, among college students, on the Narcissistic Personality Inventory. Other researchers looked at similar data and found no such rise. The most patient evidence comes from a team led by Eunike Wetzel, which followed a cohort of Berkeley students from the age of eighteen to forty-one. Most became markedly less narcissistic as they aged; only about three per cent grew more so. One of her co-authors, Emily Grijalva, has offered an explanation for why every generation nonetheless believes the young are uniquely self-absorbed: the old measure the young against who they are now, having forgotten who they used to be.
Tzani and Ioannou sensibly decline to pick a side. Their claim is narrower, and cleverer. It does not require narcissism to be spreading. It requires only that people who already have the traits become louder — more rewarded for expressing them, less often contradicted. It would even explain why the epidemic feels real whether or not it is. The behavior seems to be everywhere because the loop rewards it and puts it on display. There is an irony here the authors don't dwell on. Among the most reliably engaging content on any platform is the genre I began with: videos that teach viewers to spot the narcissists in their lives. It is grievance content with a clinical vocabulary, the very material the theory predicts a narcissistic viewer would find irresistible. The cultural conversation about narcissism may be one of the things the loop amplifies best.
The trouble is finding out whether any of it is true. Algorithmic Trait Amplification is, for now, a theory assembled from other theories — about personality, about engagement-driven design, about how therapy works — and it has not been tested against the thing it describes. The traits involved are subclinical and mostly self-reported. The data that could show one person's feed narrowing around them over months sits on the servers of a few companies that rarely share it. Researchers can build sock-puppet accounts and watch what happens to them, but a sock puppet has no personality to amplify. The most interesting question raised by the paper may not be whether the theory is right but whether anyone outside the companies could ever find out.
Where the theory has the most force is not in the population statistics but in the consulting room. Treating people with pronounced narcissistic traits is widely regarded as among the hardest work in psychotherapy. Honest feedback is often experienced as an attack, and progress depends on challenge introduced slowly, in carefully measured doses, across many sessions. Tzani and Ioannou cite estimates that roughly two-thirds of people diagnosed with narcissistic personality disorder drop out of therapy early. For those who stay, the arithmetic is grim. A therapist might get an hour a week. The phone, trained on years of the patient's own reactions, gets every other hour, and it never asks a difficult question.
Seen this way, therapy is a form of purchased friction: a room where someone is paid to tell you what you would rather not hear. The feed is friction removed as a matter of engineering, a product whose success is measured by how smoothly it gives us more of what we already want. You don't have to accept every stage of the Huddersfield model to recognize that the two are working against each other, or that one of them has a considerable advantage in hours. The authors suggest the implications reach into clinical practice, platform regulation, and how young people are taught to understand the systems that are, in a modest but persistent way, getting to know them.
In Ovid's telling, the pool never changes. Narcissus does. He lingers at the water's edge until he wastes away, and when the mourners come for his body they find only a flower. The old myth was a warning about looking too long at yourself. The new version adds a complication Ovid had no reason to imagine: a reflection that studies you in return, and gets a little better, every time you look, at keeping you there.