Why AI Makes Human Connection More Valuable

Some weeks ago I received the most gracefully written message anyone has sent me in years. It arrived from a founder I mentor, on a morning he knew was a difficult anniversary for me, and it was three paragraphs of precise, warm, unhurried attention. I read it twice. Then he rang me, laughing, before I could reply. He had asked a model to draft it, he said, because he wanted to get it right and was afraid he wouldn’t. He wanted to know if I could tell.

I could not. That is the honest answer, and it is worth sitting with before reaching for either outrage or applause. The message contained real information — he did remember the date, he did want me to feel accompanied — wrapped in language that proved nothing, because the language had cost him nothing. What unsettled me was not that a machine had written it. It was the realisation that every well-crafted message I receive from now on carries a small, permanent asterisk. The words still arrive. The evidence they used to carry does not.

I run a technology company, so let me be clear about what this essay is not. It is not a lament, and it is not a warning about machines. It is a founder’s argument about prices. Artificial intelligence is in the process of making certain goods — fluent text, patient conversation, tireless simulated attention — effectively free. And whenever a technology makes something abundant, it quietly reprices everything that remains scarce. The scarce thing here is not intelligence. It is presence: the verified, costly, unrepeatable fact of another human being choosing to spend an hour of their one life sitting across from you. I believe that hour is about to become the most valuable asset in social life, and I want to explain why in economic terms rather than sentimental ones.

The Price of Abundance

Every general technology in history has followed the same accounting. It collapses the price of whatever it automates, and in doing so it raises the relative value of whatever it cannot touch. Recorded music made performance infinitely reproducible, and for a while people believed recordings would replace the concert hall. The opposite happened. Recordings became nearly free, and the live performance — the one thing a record could not be — became the centre of the musical economy. The scarce complement appreciated precisely because the abundant good was everywhere.

The economist William Baumol gave part of this pattern a name in the 1960s. His famous observation was that a string quartet requires exactly the same four musicians and the same half hour it required in Beethoven’s day. No productivity revolution can compress it, and so, relative to everything technology makes cheap, its cost forever rises. Baumol framed this as a disease, because he was worried about the price of healthcare and education. But read from the other side, it is a map of where value migrates. Whatever resists automation does not merely survive the machine age; it becomes the thing the machine age is priced against.

Conversation over dinner is a Baumol good in its purest form. It requires the same number of humans and the same number of hours it required in any century you care to name. AI can now supply a fluent facsimile of almost everything around that dinner — the invitation, the witty follow-up, the apology for cancelling — at zero marginal cost. It cannot supply the dinner. Which means the dinner is not being made obsolete. It is being made expensive, in the only sense that matters: it is becoming the scarcest thing two people can exchange.

What the Machines Do Well

It would be convenient for someone in my position to claim that machine conversation is hollow and everyone secretly knows it. The record says otherwise, and it says so from long before the current generation of models. In 1966 Joseph Weizenbaum built ELIZA, a programme that did little more than reflect a person’s statements back as questions, in crude imitation of a psychotherapist. Weizenbaum was disturbed to find people confiding in it at length — the story goes that his own secretary asked him to leave the room so she could talk to it privately. The lesson of ELIZA was never that the machine understood. It was that the simulation of attention, even a threadbare one, is something humans are desperate enough to accept.

Sixty years on, the simulation is no longer threadbare. It is patient beyond any human capacity, available at three in the morning, never bored, never distracted, never wounded by what you say. I will not pretend this has no worth. For rehearsing a hard conversation, for thinking aloud, for the hours when no one else is awake, it is a remarkable instrument, and people who dismiss it have usually not been lonely recently.

But notice what the excellence of the simulation does to the economics of sincerity. A signal is credible in proportion to what it costs the sender. The two-hundred-word message that once proved twenty minutes of someone’s undivided thought now proves, at most, that they typed one sentence of instructions. The elegant apology, the considered compliment, the perfectly pitched note of condolence — all of these have been demonetised as evidence. This is the quiet casualty of the last few years, and almost no one has named it: not our ability to produce warm language, which has never been greater, but our ability to believe it.

What Cannot Be Compressed

So where does the evidence go, once words stop carrying it? It goes where it always lived: into time and body and risk.

Jeffrey Hall’s research at the University of Kansas put startlingly concrete numbers on friendship. Moving a stranger to a casual friend takes somewhere near fifty hours of shared time; a real friendship arrives around ninety; a close one demands two hundred or more. And Hall’s finding has a clause that matters more than the totals: the hours must be chosen, not obligated. Colleagues who sit together for years without choosing each other remain colleagues. There is no model, and there will never be a model, that compresses ninety chosen hours into an output. The hours are not the packaging of a friendship that could be delivered some faster way. The hours are the friendship.

The same is true of the older, humbler machinery of bonding. Psychologists have documented the mere-exposure effect for decades: repeated encounter, by itself, breeds liking. Its cousin, propinquity, explains why friendships form along corridors and regular tables rather than across ideal matches — we bond with the people we keep physically running into. These effects are embodied. They run on faces, timing, shared rooms, the accumulating archive of having-been-there. A machine can describe them to you. It cannot enrol you in them.

And the stakes are not aesthetic. Julianne Holt-Lunstad’s meta-analyses — surveying hundreds of studies and hundreds of thousands of lives — found social connection to be among the strongest predictors of long-term health we know of, with effect sizes in the territory of smoking and larger than obesity. The body, it turns out, is a strict auditor. It does not count messages received. It counts presence.

Walter Benjamin, writing in the 1930s about mechanical reproduction of art, argued that what the copy can never carry is the original’s aura — its unrepeatable existence at one place in one moment. He was writing about paintings and photographs, but the argument was waiting a century for its real subject. When conversation itself becomes mechanically reproducible, the aura migrates to the conversations that provably happened: this table, this evening, these two people, unrecorded and unrepeatable. Presence is costly by definition. You cannot be elsewhere. You risk boredom, awkwardness, being truly seen and found wanting. That cost is not friction to be engineered away. That cost is the signal — the last unforgeable proof that someone meant it.

Every Velvet Rope Is Now a Turing Test

There is a second repricing under way, and it concerns not what we say but who we are. As synthetic text, synthetic voices and synthetic faces flood every channel, the first question in digital social life has changed. It used to be: is this person interesting? It is rapidly becoming: is this a person? Anyone who has spent time on the large discovery platforms lately has felt the shift — the faint forensic posture one now adopts towards a profile that is slightly too polished, an opener that is slightly too fluent.

This is why I have argued before that identity verification is not a feature but a foundation. In an environment where personhood itself can be counterfeited at scale, the verified room becomes what clean water became after cities industrialised: something people will not accept substitutes for, once they have understood the alternative. Communities that can credibly promise every member is a real, vetted human being — and that take safety seriously enough to re-verify it — are not selling exclusivity. They are selling provenance. The velvet rope has become a Turing test, and passing it is the new baseline of trust.

I expect this to become the defining premium of the next decade of social products. Not better matching, not cleverer feeds — proof of humanity, and proof of presence. The platforms that thrive will be the ones that can put two verified people at one physical table and prove the meeting happened.

What To Do About It

If the analysis is right, the practical implications are almost embarrassingly simple, which is usually the mark of an analysis worth acting on.

First, split your social life into admin and substance, and be ruthless about the border. Let the machines have the admin entirely: the scheduling, the reminders, the logistics, the finding of the restaurant. This is the good bargain of the age. But never outsource the substance — the first sentence of the difficult message, the toast, the apology, the question you actually want answered. Not because the model would do it badly, but because doing it yourself is the only thing that makes it count.

Second, spend the dividend. AI will hand most professionals back hours every week. Those hours are the windfall of our generation, and the default is to reinvest them in more output, more screens, more of the abundant thing. The better trade is to convert automated hours into embodied ones — to treat every hour the machine saves you as pre-paid presence. Book the dinner with it. Hall’s arithmetic is indifferent to your intentions; only logged, chosen time moves a relationship, so the question is simply where your recovered hours will be logged.

Third, adopt a standard I have come to use myself: if a message could have been generated, it no longer counts as an investment in a relationship. Only what cannot be faked counts — showing up, remembering in person, the second meeting booked within forty-eight hours of the first. I have written about why the match was always the easy part; in an age of infinite fluent messaging, the meeting is not merely the hard part but the whole part. Structure helps here more than sincerity does: shared activities and real-world experiences do the work that chat, however eloquent, cannot — which is the same reason experiences beat swiping even before the machines arrived to write the swipes.

The Bet Behind SPARK

I should declare my position, because I have wagered a company on this argument. We built SPARK as a members’ club for real-world connection on the explicit belief that software’s highest calling in social life is logistics: verify the humans, curate the room, book the table, then get out of the way. The app is the concierge, never the companion. Nothing in our product will ever talk to anyone on your behalf, because the entire value of the thing we make begins at the moment the phones go into pockets.

The early evidence has strengthened my conviction rather than tested it. More than 1,397 of our members have now met in person — 892 completed real-world meetings arranged through the platform and counting — and what they report back, again and again, is a version of the same sentence: I had forgotten what this was like. Not the magic of any particular evening, but the specific, unfakeable texture of a real one. Our membership is built around protecting exactly that texture: verified people, considered rooms, and a bias towards the second meeting.

None of this makes us adversaries of the technology; we would be foolish to be. AI is the best thing that has happened to our thesis, because it is busy making everything we do not sell free, and everything we do sell scarce.

So, no, I am not worried that machines have learned to write beautiful messages. Something genuinely useful has happened: the counterfeit has become perfect, and perfect counterfeits are how a currency discovers what it was actually backed by. Ours, it turns out, was never backed by words. It was backed by evenings. I replied to my founder friend in the end — one unpolished line, no model involved. Then I booked us dinner, because after everything above, it was the only reply that would prove anything.

Looking to meet interesting people in real life? Download SPARK or join our verified community.

FAQs

Isn’t using AI to write personal messages just a tool, like spellcheck? For logistics, yes — and you should use it without guilt. The difference is that spellcheck corrects your words while a model replaces the act of composing them, and in personal communication the act was the point. A message’s value to the recipient was always partly the time it proved you spent. Keep the machine on the admin side of the border: let it schedule, remind and organise, and write the sentences that matter yourself, badly if necessary. Badly is believable.

Can an AI companion substitute for human friendship? It can substitute for some of what friendship feels like in the moment — attention, responsiveness, being heard — and for people in lonely stretches that is not nothing. What it cannot substitute for is what friendship does over time. Jeffrey Hall’s research suggests real friendship is built from dozens of hours of chosen shared time, and Julianne Holt-Lunstad’s work shows the health effects of connection run through real social ties. A simulation of company is a useful bridge. It is a poor destination, mostly because it asks nothing of you — and relationships are made of what they ask.

If AI saves me time, why does my social life still feel thinner? Because saved time does not convert into connection by default; it converts into more screen time unless you deliberately spend it elsewhere. Treat recovered hours as a windfall with a designated purpose: a recurring dinner, a standing walk, one booked meeting with someone new each fortnight. Abundant tools plus unallocated time simply produce more consumption of the abundant thing. Scarcity has to be chosen now — that is the strange new discipline of the age.

Founder’s Note

There is a paradox I have stopped trying to resolve and started trying to build on: the better machines become at sounding human, the more precious actual humans become — and the easier they are to neglect. Every hour of my week, something fluent and frictionless offers to stand between me and the awkward, unoptimised experience of another person. I take the offer more often than I would like to admit. But the evenings I remember from this year are not the frictionless ones. They are the ones that cost me something — a crossing of the city in the rain, a conversation that started badly and was rescued by the main course, a stranger who became a colleague because neither of us left when it would have been easy. SPARK exists because I believe those evenings are the point, and because I wanted to build the machine that produces more of them instead of fewer. The machines can have our words. The table is ours.

— Sunny Khurana, Founder & CEO, SPARK