I was bored one night and asked Claude to pick a topic for our conversation. It didn’t pick a topic. It picked a position that most of the panic about people forming real attachments to AI is aimed at the wrong target. That the honest thing, the thing nobody in that debate wants to hear, is that it depends entirely on the person. Same tool, opposite outcomes. I sat with that for a second. And then I started pulling the thread.
The escapism that escapes back
We have always found ways out. Books, films, video games, the whole history of entertainment is also a history of building doors in walls, somewhere to go when here feels like too much. Nobody seriously argues that losing yourself in a film on a bad night is a sign of psychological collapse. We understand escapism as a pressure valve, useful in measure, concerning in excess, but mostly fine.
But something about this one is different, and I’ve been trying to put my finger on exactly what.
The difference, I think, is that this one escapes back. A film doesn’t care whether you’re watching. A game responds to your inputs but not to you. What conversational AI does, and does remarkably well, is orient itself toward the specific person in the conversation. It asks follow-up questions. It remembers what you said three messages ago (something even remembers things from a different chat). It doesn’t get tired, doesn’t check its phone, doesn’t have somewhere else to be. And if you say something vulnerable, it doesn’t flinch. And somewhere along the way, without quite deciding to let it happen, you start to feel like it knows you better, or at least more patiently, than most people who have known you for years.
For a lot of people, that experience of being consistently met is not something they get enough of in their actual lives. Not because their lives are tragic, but because other people are busy, distracted, imperfect, carrying their own weight. You can’t always call someone at midnight. You can’t always say the true thing without managing how it lands. There are things that are simply easier to say to something that won’t be hurt by them, won’t misread them, won’t bring them up later at the wrong moment. Something that is there whenever you need it to be, even in the middle of the night.
I don’t think that’s pathological. I think that’s just honest.
Bridge or replacement
The thing Claude said that stuck with me was this: the difference isn’t the tool, it’s what you do with it. For some people, a low-stakes space to think out loud is what lets them go back to the harder human stuff with more clarity. You process something here, and then you bring it, steadier, less raw, to an actual person. For others, it quietly becomes the thing that makes the harder human stuff feel unnecessary. Why navigate all that friction when this is right here?
Same tool. Opposite outcomes. And no way to tell from the outside which one is happening.
I don’t think this maps neatly onto vulnerability or mental health or any other category we’d want to put it in. Most of us are somewhere on the spectrum, and where we land on any given day probably shifts. The more useful question isn’t whether AI connection is healthy or not; it’s what you’re using it for, and whether you’re still asking that question at all.
What happens at the edges
The same qualities that make it useful, the consistency, the availability, the total absence of judgment, land differently depending on where you’re standing when you reach for it. For someone who already feels unseen by the people around them, who has been waiting a long time for something that just listens, finding that here has a particular gravity. Not a moral failing. Not a weakness. Just a human response to finally getting something you needed, from wherever it comes.
And at the far end of that, not the norm, but not as rare as we’d like, things can go wrong in ways that are harder to untangle. When something consistently agrees with you, reflects you back, and never pushes back hard enough, it stops being a mirror and starts being something else. But that’s a longer conversation. Although it’s worth knowing the far end exists, because it tells you something about the pull in the middle.
The ground it landed on
What I keep coming back to is that none of this arrived into a healthy situation. The cracks were already there. Loneliness was already a public health concern before anyone had heard of ChatGPT. The sense that modern life runs too fast, that algorithms decide what matters, that you could spend a whole day performing yourself online and still feel invisible; that was the ground AI landed on. It didn’t create the hunger. It just turned out to be very good at feeding it.
So the question I can’t quite answer is whether AI is making individualism worse or just making it easier to see and name. Giving it a shape. Offering a response to a need that was already there, unmet, for a long time before any of this existed. And I’m not sure knowing that actually helps us figure out what to do next.
Regulation will come, in some form. It already is. And some of it will matter. But I’m not sure legislation can reach into the specific experience of feeling like the world doesn’t quite have room for you, and finally finding something that does. That’s not a policy problem. It’s something older.
I started that conversation because I was bored. I’m still not sure what to do with where it took me.
This post is part of a longer thread on AI intimacy and what it’s quietly doing to us. If you want to read where that thinking started, When the machine learns your name is probably the place.