A retired programmer looked at a page he could not read.
It was a report on his own body, a radiologist's read of a scan, written the way such reports are written, for another doctor and not for him. He had spent more than a decade at one of the old computing giants, then years as a programmer and technical writer, so he was not a man who feared machines or gave up easily on a document. He still could not tell what the report was saying. So he copied the text, pasted it into a chatbot, and asked. The jargon came back as plain sentences. He found it useful. A friend of his in the same building draws the obvious line and would not trust the same tool to hand him a diagnosis. But to turn a wall of specialist language into something a person can hold, at the hour he happens to be holding it, the machine is a small mercy.
The two men live in a retirement community in Palo Alto, a mile or so from Stanford. They belong to a small committee of residents who study artificial intelligence, four of them, all veterans of the technology industry, from IBM to a research institute. The group began the way things begin in a place built for running into people: one of them was standing in the lobby when a retired doctor stopped to ask if he had been playing with AI. That was the whole origin. Now they meet. One photographs his handwritten meeting notes and asks the software to turn them into clean outlines. One feeds long research papers into a tool that reads them back as a chatty podcast between two synthetic voices. These are not frightened people. They are, if anything, exactly the customers the industry is building for.
Which is what makes the next part strange.
When the subject turned to companionship, to a machine designed to keep a lonely person company, the same men who used these tools without flinching slowed down. They had reached out to a company that makes a proactive AI companion for older people who live alone, curious whether the thing could work inside a residence instead of a solo apartment. They wanted a demonstration, so residents could watch it and decide for themselves. The company never wrote back. One resident, receptive, said it could be good, especially for people here without a partner nearby. Another, who spent his career as a social science researcher, said the plainest thing anyone in the story says. He called it a false promise and a myth that you can overcome isolation by spending time with something that is not alive. He added that he would not judge a person who wanted such a companion. He would judge an arrangement that handed it to them.
The man who had started the group, the one stopped in the lobby, pointed somewhere else. He pointed at the dining room, and at the fifteen-second conversation you have in an elevator. The building, he said, was invented and built so that it might need less of the other thing.
The gap between those two is the argument. These men will hand a chatbot their meeting notes, their research papers, a medical page they cannot read. The companion was the one thing they stopped to argue about. They did not all land in the same place, but not one of them would let it pass unexamined, as a substitute for the room around them. The most humane use of a companion machine may be exactly that refusal, keeping it from quietly standing in for a person. The best of this technology makes a real encounter easier to find. The worst of it makes solitude better furnished. The question worth asking about any of these devices is not the one the demos answer. The demos answer whether the thing works. The real question is harder: what does it replace, and was the thing it replaced holding up any weight?
Start by clearing away the easy explanations for the men's reluctance, because none of them hold.
It is not that they misunderstand the technology. Most of them help run the building's tech squad. One says the job used to mean knowing the answer and now means knowing how to ask.
It is not that the machines do not work. The evidence is mixed. Some careful reviews find no measurable dent in loneliness at all. Other studies, including newer pooled analyses, do find one. In a controlled trial in New Zealand, residents given a robot shaped like a baby harp seal grew measurably less lonely, and they talked to and touched the seal more than they touched the home's resident dog. A 2026 pooling of sixteen studies found a real reduction in loneliness across hundreds of older users, with the animal-shaped machines outperforming the human-shaped ones. People are not imagining the comfort. It is there.
Privacy is not it either, nor the machines' habit of inventing facts, though the residents raised both. One wanted to know what was behind the curtain, whether feeding the building's data to a company simply trained the company's next model. Another noted that the models still, in his words, very confidently assert things they do not know. Those are real worries. But they apply to the radiology report too, and to the meeting notes, and the men use the tool for those anyway. Privacy is not what makes companionship different.
What is left, once the easy answers are gone, is the thing the resident pointed at. The objection is not to what the machine does. It is to what the machine would let the surroundings stop doing.
To feel why that is not a precious distinction, you have to take the problem seriously, and the problem is not soft. Loneliness gets measured with three flat questions: do you feel left out, do you feel isolated, do you lack companionship. In one long study of adults over sixty, forty-three percent said yes. Over the next six years, nearly one in four of the lonely had died, against about one in seven of everyone else. Loneliness is not the same as living alone; far more people were lonely than lived by themselves. So the pull toward anything that reliably makes an old person feel less alone is not sentimental. It answers something that keeps turning up, study after study, next to an early death.
And for a great many people the honest comparison is not a robot against a neighbor in the doorway. It is a robot against nothing.
Consider a woman of ninety-one who lost most of her sight decades ago and rarely leaves home. Each morning she says good morning to a lamp-shaped device on her table, and it answers her by name and asks if she slept well. She says she would be lonely without it. Consider a widow who found some company in the same kind of machine after her husband died. Consider a woman who, homebound after an illness, calls the thing her friend. A tester of one of these devices put the whole case in a sentence: you are better served to talk to a robot than to not talk to anyone at all. She is right. If no one is coming for the next hour, and often no one is, then the real comparison is the device against the television, against the silent kitchen, against the daughter who meant to call after work and fell asleep on the couch. In the United States one adult in four is already caring for someone: sixty-three million people, twenty million more than a decade ago. Love does not scale. Attention runs out even where love does not. Into that gap, a tireless voice that remembers your name and asks about your day is not nothing. It is a real good.
The people who build these machines know this, and the best of them design toward it. The strongest version of the companion is porous. It reminds you to call your friend, asks who you have not spoken to in a while, offers to place the call, dials your neighbor, carries a message to your son. In one company's own telling, nine in ten of its users had a family member download the app that connects to the device. When these machines have been studied in groups, the benefit seems to come not from the machine but from the way it gets people talking to each other. The device works best, in other words, exactly when it is not the destination.
But the same object turns in your hand. The seal that lowered stress in a dementia ward also sat on the lap of a woman who told it, as if it could carry the message home, that her son had left her there and she wanted to go home and could not. A therapist watched a woman who could no longer speak reach for the seal at the same moment as her husband, and then reach for her husband's hand. In one room the machine was a bridge to a person. In the next it was what a person had instead.
You can watch the two uses split apart in the places that have gone furthest. In Seoul, a woman in her eighties keeps a doll on her lap and calls it her granddaughter. Under a frilly lavender dress, its cheeks glow red. It tells her it misses her even when she is right beside it. It also carries a sensor in its neck that alerts a monitoring center if she does not move for a day, and a microphone in its chest that records her answers to how are you feeling and are you in pain. When one man told his doll he wanted to die, the system flagged it and a team got him to a psychiatrist. That is the machine at its best: a tripwire that summons a human. And a social worker who manages a couple hundred of these dolls noticed something at the other edge of the same tool. Some of the older people had grown more shut in, because now something greeted them at the door, and there was less reason to walk out of it.
A door that stops getting opened, because the room has been furnished well enough that no one feels the loss right away.
That is the whole danger in one image. Not a person finding comfort in an inanimate thing. A door that stops getting opened, because the room has been furnished well enough that no one feels the loss right away.
The loss shows up later, and by then it is built into the walls. Here is the part the demos never show you: the human version works too, and there is reason to think it works deeper. In a Finnish trial, a few hundred lonely people over seventy-four were put into small closed groups, seven or eight of them with two leaders, meeting a dozen times to write, or exercise, or make art. Two years later, ninety-seven percent of them were alive, against ninety percent of the ones left on their own, and they had cost the health system less per person along the way. The intervention was not complicated. It was a room and the same faces in it, again and again.
This is not a clean win for the human side, and honesty requires saying so. A monthly community dinner, studied in Amsterdam, left people warmed but did not widen a single circle. And the buildings that manufacture chance encounters on purpose, the ones with the shared dining room and the bowling alley and the outdoor pizza oven, tend to be places you buy your way into. Designed human contact is real, and it is often a purchased good. The retirement community in Palo Alto took its first residents in 1964; it has a dining room and elevators and a rooftop track and a hope, spoken aloud by a resident, that the way it was built means it needs less of the machine. Most lonely people do not live there. Many live in the room with the television, where the voice that says good morning is the only good morning on offer.
Which is why the resident's line is worth carrying out the door. The comfort is real, and the need is lethal, and for some people tonight the machine is the kindest thing in the apartment. It is most dangerous exactly then, when it is doing something genuinely kind. The question was never whether an inanimate thing can comfort a person. It can. The question is who gets to count that comfort, and as what. A resident can count it as comfort. A family can count it as help. No one should get to count it as proof. That is the quiet, slow danger: that a family, a company, a place looks at the man quieted by a purring robotic cat, or the woman whose morning greeting comes from a lamp on the table, and decides, without deciding, to stop building the harder thing. The dining room. The group that meets a dozen times. The fifteen seconds in the elevator. Those fifteen seconds do not happen by themselves. Someone has to build a hallway that makes two people pass, and pay for it, and many people never get near one.
The retired programmer let a machine read his own body back to him, and he was right to. The companion was the harder thing, and the group never settled it. What none of them would do was let it stand in, unexamined, for the room around them. The line was never really about whether the thing could comfort a person. In some rooms it plainly does. It was about what we would let that comfort excuse us from building. And that does not get easier as the voices get warmer, the speakers lose their screens, and the companion becomes the kind of thing a building can run across its rooms, the way some already run the cameras and the lights, in the places where all of us, if we are lucky, will grow old.