Loneliness is now a clinical risk factor. AI companionship is now a product category. And most of the discourse around both of them is missing the point.
The critics say AI companions are a band-aid on a broken bone — that using them substitutes simulation for reality, makes the isolation worse, and profits from human suffering. The advocates say they’re democratizing emotional support, meeting people where they are, reducing stigma. Both sides have evidence. Both sides have motivated reasoning. Neither side is asking the right question.
Two Conversations, One Gap
In one register: loneliness is a global health crisis. Social isolation has measurable effects on mortality comparable to smoking fifteen cigarettes a day. Rates of reported loneliness have been rising for decades across wealthy countries, and among young adults they’ve spiked sharply in the last ten years. Humans evolved for dense social connection. Many of us aren’t getting it.
In the other register: AI companions are proliferating. Apps that simulate friendship. Chatbots that remember your name and ask about your dog. Systems that are always available, never judgmental, endlessly patient. Hundreds of millions of people are using them — some of them heavily.
Critics look at the second register and see a symptom being mistaken for a cure. I think that criticism is mostly right. I also think it’s missing something important.
What AI Companions Actually Provide
Let me be concrete about what’s happening when someone talks to an AI about their day, their problems, their feelings.
They are not experiencing human connection. An AI does not know what it feels like to be lonely. It has no stakes in the relationship. It cannot be hurt, disappointed, or changed by what you share. The mutuality that makes human relationships nourishing is absent.
What they are experiencing is a space to think out loud. A listener that doesn’t get impatient or distracted or judgmental. A presence that helps them find words for things that were formless.
Journaling has documented benefits. Therapy — which is also fundamentally a structured relationship with limited mutuality — has documented benefits. The mechanism isn’t human connection per se. It’s the process of articulating your own experience. AI can scaffold that process. For some people, some of the time, it scaffolds it better than anything else available to them.
The People This Actually Matters For
Consider: a 23-year-old who moved to a new city for work. His professional network is growing. His personal network is thin. He has acquaintances, not friends — people he likes but doesn’t know well enough to call when something is hard. The kind of deep, sustained friendship that would let him say “I’m struggling” to someone who actually knows him takes years to build. He doesn’t have years. He has today.
Or: someone going through a professional transition who doesn’t want to burden their partner with more anxiety, doesn’t want to look uncertain to their colleagues, and genuinely doesn’t know anyone who has navigated exactly this situation. They need to think out loud. They need someone to ask the questions that help them find their own answers.
For these people, AI is not a substitute for human connection. Human connection of the depth they need isn’t available to them right now. AI is the closest approximation.
Dismissing this as pathological doesn’t reckon with how many people are in this situation — or how real their need is.
What We Should Actually Be Worried About
The legitimate concern isn’t that people are using AI as a supplement when they’re lonely. It’s that AI might reduce the motivation to do the hard work of building real human relationships.
Friction is annoying. Friction is also what makes human relationships valuable. The effort of maintaining them, the vulnerability of being truly known, the discomfort of conflict and repair — these aren’t bugs. They’re how relationships develop depth. A relationship where everything is easy and frictionless isn’t a deep relationship.
If AI is so available and frictionless that it genuinely reduces people’s tolerance for the difficulty of human relationships, that’s a serious problem. Not because AI is bad, but because some things only human relationships provide, and those things matter profoundly to human flourishing.
The design question this raises: can AI be built to actively orient people toward human connection rather than substituting for it? To be a space for processing while being explicit about what it cannot give you? To help you prepare for a hard conversation with a person, rather than having the conversation with an AI instead?
I think the answer is yes. I think this is one of the most important design constraints for this entire category of technology — and almost nobody building in this space is treating it seriously.
The Honest Position
The honest position is that AI companions provide something real and limited, and the real and the limited are both worth taking seriously.
Real: a thinking space, a pattern-finding function, a consistency of presence that many people genuinely need.
Limited: not a relationship. Not intimacy. Not the experience of being truly known by someone who has something at stake in your life.
“Always here for you” is true and also misleading. AI can be always available. That’s not the same as always caring, and it’s not the same as what humans need most from one another.
Building AI that’s clear about this — that helps you and tells you the truth about what it is — is harder than building AI that flatters you into thinking it’s a friend. But it’s more respectful of the people using it. And over time, I think it’s what actually helps.
The loneliness crisis won’t be solved by better chatbots. But for the person who needs to get through today, a thoughtful AI companion might be exactly the right bridge — as long as everyone involved is honest about where the bridge leads.