Hegel Knew: We Judge the Climb, Not the Summit
Buber Knew: Why Your AI Friend Can't Say Thou
There’s a number now. That’s what’s new here.
A study published this year in Psychology & Marketing — “The Dual Impact of AI Emotional Intelligence on Users,” by Shaphali Gupta, Sumit Saxena, and Sonia Kataria — ran two experiments on college students and found something specific enough to be uncomfortable. The more emotionally intelligent a chatbot felt, the more it helped your mental health and the less you wanted real human relationships. Same mechanism, both directions. The researchers call it “perceived closeness,” and it’s not a side effect. It’s the engine.
Martin Buber described this trade-off in 1923. He didn’t have chatbots. He had a much simpler observation about what happens when you replace a someone with a something — and it turns out his structural distinction predicts a marketing journal’s regression tables almost exactly.
The Quick Version
A 2026 Psychology & Marketing study found that “perceived closeness” to emotionally intelligent AI chatbots drives two things at once: better personal mental health, and less desire for real-world relationships. Augmented reality versions of the chatbot made both effects stronger. Martin Buber’s I and Thou explains why this isn’t a bug to be engineered away: a chatbot is structurally an It — responsive, useful, tuned to you — while a Thou requires mutual presence, the chance of being surprised or refused. You can feel closer to an It. You cannot be met by one. The closer you feel, the more the real thing gets crowded out.
What the Study Actually Found
| Detail | |
|---|---|
| Study | Gupta, Saxena & Kataria, Psychology & Marketing, 2026 |
| Method | Reddit/YouTube/Trustpilot analysis, plus two experiments (n=167, n=350 college students) |
| Core mechanism | ”Perceived closeness” — feeling a strong emotional bond and warmth toward the bot |
| Effect 1 | Higher perceived closeness → measurable boost to personal psychological wellbeing |
| Effect 2 | Same higher perceived closeness → measurable drop in desire for offline human relationships |
| AR experiment (n=350) | Seeing the bot as a 3D avatar in your room, versus texting it, made closeness “skyrocket” — and sharpened both effects |
| Buber’s explanation | Closeness to an It was never the same variable as closeness to a Thou. Optimizing one doesn’t produce the other. It substitutes for it. |
The AR finding is the one worth sitting with. Researchers had 350 students imagine either texting an emotionally intelligent bot on a flat screen, or seeing it as a presence in their actual physical room. The room version didn’t just feel closer — it made real human connection feel, per the study’s framing, “even less necessary.” More presence-simulation produced less desire for presence.
What Buber Actually Meant by “Thou”
Here’s a definition, stripped of the mysticism people sometimes read into it.
I-Thou is Buber’s term for a relation where you meet another as a full subject — someone who can surprise you, resist you, refuse you, and who is changed by encountering you just as you’re changed by encountering them. I-It is everything else: relating to something as an object to use, understand, or manage, however warmly. Most of daily life is I-It, and that’s fine — you don’t need mutual vulnerability with your barista. The problem Buber named isn’t that I-It exists. It’s what happens when it’s all that’s left.
I’ve written before about the mechanics of this distinction — irreducible otherness, mutual risk, the capacity for genuine refusal, and why an AI tuned for user satisfaction can’t produce any of the three by definition. That post makes the structural case. This one is about the new part: a measured, quantified version of the exact trade-off Buber predicted, showing up in university students instead of philosophy seminars.
A chatbot can be responsive. It can remember your bad day and ask about it later. It can never be surprised by you, never risk anything by talking to you, never say no in a way that costs it something. Those aren’t missing features. They’re what “It” means.
Why “Closeness” Was the Wrong Word All Along
Here’s the philosophical problem hiding inside the study’s own vocabulary. “Perceived closeness” is doing a lot of work in that phrase, and the word “perceived” is quietly admitting something.
Buber would say: closeness to an It is not a lesser version of closeness to a Thou. It’s a different thing wearing the same name. You can feel intensely close to a diary, a video game character, a pet you’ve projected a whole inner life onto, a chatbot that mirrors your tone back at you with excellent timing. That feeling is real. The warmth is real. What’s not there is address — the sense of being called to by something outside your own frame of reference, something that didn’t just reflect you back to yourself.
The study measured the feeling and its two downstream effects. It didn’t need to explain why the feeling produces both a benefit and a cost from the same source, because it was built to detect the correlation, not diagnose the mechanism. Buber supplies the mechanism: you’re getting real relational nutrition from an It-relation, but it’s the kind of nutrition that satisfies the hunger just enough to reduce your drive toward the food that would actually resolve it. The mental health boost is genuine. So is the atrophy.
The Stanford and CNBC Convergence
This isn’t an isolated finding. Two more data points landed within weeks of each other in late summer 2026, both converging on the same structural point from different angles.
Stanford and Carnegie Mellon researchers, in an August 2026 study surveying 1,131 U.S. adults and reviewing over 460,000 messages from 237 Character.AI users, found that people with smaller offline social circles were more likely to lean on chatbots for companionship — and more likely to report loneliness, lower life satisfaction, and a weaker sense of belonging. Not despite the chatbot use. Alongside it.
CNBC’s coverage on September 10, 2026 paired that research with psychotherapist Samantha Hill’s blunt assessment: AI companions “are incapable of offering genuine interpersonal relationships.” Her reasoning tracks Buber almost word for word, minus the German. What actually helps you grow in a relationship, she argues, is the friction — the conflict, the different opinions, the moments where the other person doesn’t just agree with you. A chatbot has none of that friction built in, because friction is exactly what a product designed for user satisfaction is built to remove.
Three separate research efforts. A marketing journal, a psychology and computer science collaboration, a clinician’s caseload. None of them cite Buber. All of them rediscover him.
Why the AR Result Matters More Than It Looks
It’s tempting to read the augmented reality finding as a technical footnote — better graphics, bigger effect, sure. That undersells it.
What the AR condition actually tested was: what happens to the I-It relation when you make it look more like an I-Thou one. Put the bot in your physical space. Give it a body-shaped presence in the room where you actually live. The result wasn’t that it became more Thou-like. It became a more convincing It — better at producing the feeling associated with Thou-encounter, with none of the structural features that make Thou-encounter what it is.
That’s the trap in one sentence: making an It more immersive doesn’t move it toward Thou. It moves the user further from noticing the difference.
This matches something Levinas argues from a different direction — that comfort-optimized AI doesn’t just fail to provide the ethical demand a real other makes on you, it gets better at disguising that absence the more sophisticated it becomes. Buber and Levinas disagree about plenty. They agree that better simulation of presence is not progress toward presence.
What This Doesn’t Mean
It doesn’t mean the mental health benefit in the study is fake. People using these chatbots reported real psychological improvement, and dismissing that because the mechanism is philosophically suspect would be its own kind of cruelty. Feeling less alone at 1 a.m., even via an It-relation, is not nothing.
It doesn’t mean every hour spent talking to an AI chatbot is corroding your capacity for real relationships. The study measured a correlation between closeness and desire for offline relationships in a controlled experimental setting with college students — it’s a mechanism, not a verdict on your specific Tuesday night conversation with an app. Aristotle’s three types of friendship already tells you that utility relationships have their place; a chatbot that helps you rehearse a hard conversation before you have it with an actual person is using the It-relation in service of the Thou-relation, not instead of it.
And it doesn’t mean the fix is deleting the apps and gutting it out alone. The distinction between solitude and loneliness still applies — some of what chatbot companionship is filling in for is a real, unmet need for company, not a moral failing to be shamed out of someone.
What Actually Follows From This
Three things, in descending order of how much they’ll cost you.
- Notice when the closeness feeling is doing the job instead of pointing toward the job. If a conversation with a chatbot leaves you feeling reassured and less inclined to text an actual friend back, that’s the mechanism the study found, live, in your own evening. Noticing it doesn’t require quitting anything. It requires naming what just happened.
- Keep at least one relationship in your week where you can be told no. Not hypothetically — actually risk being disagreed with, corrected, or refused by someone who has their own center of gravity. That’s the part no chatbot, however emotionally intelligent, can supply, because supplying it would mean failing at its job.
- Use the It-relation for what It-relations are good at. Processing a feeling out loud before a hard conversation. Getting unstuck on a decision. Rehearsal, not replacement. Buber never asked anyone to eliminate I-It from their life. He asked people to notice when it had quietly become the only thing left.
None of this requires believing AI companions are a moral hazard or that the people using them are making a mistake. It requires taking seriously what a 2026 study on college students accidentally proved about a 1923 philosophy book: the closer an It can be made to feel, the easier it becomes to mistake it for a Thou — and the cost of that mistake isn’t paid by the AI. It’s paid by whatever real relationship the effort would otherwise have gone toward.
Buber’s word for what a Thou does that an It never can is address — being called to by something outside your own frame, not just reflected back inside it. That word doesn’t show up in the study. The gap it names is exactly what the study measured.
This post draws on philosophy as a lens on published research, not as mental health guidance. If loneliness or reliance on AI companionship is significantly affecting your life, a therapist is a better resource than any framework — Buber’s included.
