What the machine teaches us about listening - critical summary review

People are rating responses from artificial intelligence as more caring than those from people trained to comfort others. What that says about the way we listen.

the conversation no one interrupts

It happens on millions of screens every night. Someone opens an app, types out something they have not told anyone, and gets back a calm reply, unhurried, with no wince. A young Ukrainian woman living in London described this to a reporter at a major British public broadcaster with a matter-of-factness that is a little unsettling. She knows she is talking to a program. Even so, when a relationship ended, it was the conversation with the machine that gave her room to understand what she was feeling. Her friends showed up with the verdict already formed, "he's an idiot," and closed the subject before she could open it. The program had no verdict at all. It just waited.

It may sound strange, and she herself asked not to be named. But she is far from an isolated case. In 2025, the biggest roundup of how people actually use artificial intelligence put "therapy and companionship" in first place, ahead of writing code and organizing daily life. The share tied to emotional and existential support jumped from about 17% to nearly 31% in a single year. What began as a work tool became, for a lot of people, the place where they talk about what hurts.

the finding that stings

What makes this hard to swallow came out of a lab at the University of Toronto. Across four experiments with more than 500 participants, researchers showed written responses to happy and sad situations and asked people to rate how much care each one held. Some came from ordinary people, some from responders trained in crisis listening, some from a chatbot. The machine's won. They were judged more compassionate, more attentive, better at making the other person feel understood, and were preferred in nearly seven of every 10 comparisons. The detail that drives it home: even when raters knew which response came from a program, they kept choosing the program's.

Before concluding anything about the machines, it is worth stepping back. The study's lead has a sober explanation. AI does not get tired. It does not have an off day, it is not in a rush to tell its own version, it does not feel that flicker of discomfort that makes us steer away from someone else's sadness. It answers with the same patience every time, because it was built to please and to keep being used. That is not kindness. It is the absence of everything that usually gets in the way of ours.

And that is where the story stops being about technology and starts being about us. If a program with no feeling at all beats trained people, what it gets right is exactly what human listening has been getting wrong. The good news hidden in the discomfort is that the list of things it gets right is short and not the least bit mysterious.

what the machine does that we forgot

The first is the most obvious and the rarest: letting the other person finish. We interrupt for a thousand reasons, to fill the silence, to "help" find the word, to jump ahead to the answer we think is better. Every cut, however well meant, robs the other person of the chance to work out their own thought. The machine does not cut in, so the thought arrives whole.

The second is naming what the other person feels without dressing it up. When a friend says the cat died, the reflex is to comfort fast, "he had a good life and was well loved." It sounds gentle, and it works to shut the grief down early. Language programs were built to recognize the emotion and hand it back, which makes the person feel that their sadness fits there, that it is okay to sit in it for a moment without scaring anyone off.

The third is resisting the urge to fix. A lot of people, especially those in a boss's chair or a parent's, believe their worth lies in the perfect piece of advice. But most of the time, the person venting was not asking for a solution. They were asking for company in the problem. The machine, with nothing to prove, offers presence before it offers a plan, and that is exactly what makes a person feel heard.

The fourth is not hijacking the conversation. Someone mentions a loss and we answer with our own similar loss, sure it proves we understand. The moment we start telling our story, we stop listening to theirs. The machine has no stories to tell. It does not fall into the trap because it has no "self" to put at the center. We do, which is why keeping the focus on the person speaking has to be a conscious choice.

None of this is sophisticated technique. It is what Marshall Rosenberg said all along when he defined communication as just two things, speaking and listening, and pointed out that most of the noise between people comes from the judgment baked into every sentence. The program does not judge because it has no values and no dislikes. We can choose not to judge on purpose, and it makes an enormous difference when we pull it off.

where the mirror clouds over

It would be nice to stop here, but there is a limit that has to be clear. The machine does not care. It simulates warmth drawn from billions of human conversations, and it does that very well, but there is no one on the other side feeling anything for you. As long as the point is learning from it, fine. When it becomes the only ear available, the math changes. Recent studies show that heavy, daily use is linked to more loneliness, more dependence, and less real-world contact, and that the same tool that soothes at first can deepen the emptiness once it replaces flesh-and-blood people. Frictionless comfort is habit-forming, and the world outside, with its imperfect conversations, starts to feel like too much work.

Maybe that is the mirror's best use. Not to convince us the machine is enough, but to notice how far the bar for listening has slipped without our seeing it, and how simple it is to raise it again. The next time someone starts to talk, you can try what the program does for free: stay quiet until the end, name what the person feels, hold back the urge to solve, and keep the subject off yourself. Feeling heard is still one of the rarest things one human can give another. And unlike the machine's affection, this one is real.

what to do with this information

For people who lead teams: in the next hard conversation, hold the advice. Let the person finish without cutting in, and say out loud the emotion you picked up before you propose any plan. Listening first, solutions later.

For people who follow the tech debates: notice that the shock over "AI listens better" says less about how far the machines have come and more about a long-standing thinning of human listening. It is a cultural data point, not just a technical one.

For people who just want better conversations at home: try one thing at a time. This week, only not interrupting. Next week, only not telling your own similar story. Small moves, and the effect shows up fast.