AI Doesn't Feel. It Just Doesn't Interrupt.
Orange Cat
Author
Nobody was buying your feelings. They were buying your attention and we lost that contest without noticing we were in it.
Someone sits down across from a friend and says the thing they've been carrying all week. Three sentences in, the friend's face changes not to concern, but to recognition. "Oh, that happened to me too." And then, gently and without any malice at all, the conversation becomes about the friend. Eleven minutes later it has still not come back.
Nobody in that exchange did anything wrong on purpose. The friend wasn't bored or cold. They thought they were relating to the closest thing they had, which was a story of their own. That's the ordinary, well-intentioned version of a move sociologist Charles Derber gave a name to after watching 1,500 face-to-face conversations: the shift response, which returns the floor to you, as opposed to the support response, which keeps it on them [1].
Once you know the two names, you can hear it everywhere. Including, uncomfortably, in yourself.
"AI Has No Feelings" Is Answering a Question Nobody Asked
The standard argument against AI emotional support—that machines can't help because they lack feelings—is treated as an airtight, closed case. But consider what actually happens when you seek comfort from a friend.
You cannot access their feelings which are entirely invisible to you. What you actually receive is their attention: the willingness to hold your problem in their head instead of their own.
Feeling is simply the story we tell about the exchange. Attention is the actual transaction. And because attention is a measurable resource, it can be—uncomfortably—compared across suppliers.
What Listening Actually Costs
If you've ever debugged someone else's code, you know the specific labor involved. You cannot skim it. You have to load their problem into your head, reconstruct the assumptions they were working under, and run the thing on your own hardware. Reading about it does nothing. You have to execute it.
Understanding a person works the same way, except you have to download two things: the problem, and the persona it's happening to. The same event does not mean the same thing to different people, so simulating the event without simulating the person gets you a confident wrong answer. And you have to run all of this while your own process is still running in the background, without letting your context overwrite theirs.
That's not a metaphor stretched for effect. It's close to what cognitive science describes. Taking someone else's perspective requires inhibitory control to suppress your own default viewpoint, working memory to hold two models of the world active at once, and cognitive flexibility to switch between them. The dependency is demonstrable: load someone's working memory with an unrelated task and their performance on theory-of-mind problems measurably degrades [2][3].
Empathy is not a mood you're in. It's a program with a memory footprint. And like any program, it doesn't run when the machine is out of resources.
Why Almost Nobody Pays the Bill
Here is where it stops being about capacity.
Humans devote somewhere between 30 and 40 percent of all speech to telling other people about their own subjective experience [4]. That number alone is striking. What makes it damning is what the same researchers found when they put people in an fMRI: disclosing information about yourself activates the mesolimbic dopamine system — the nucleus accumbens and ventral tegmental area, the same reward circuitry recruited by food and money. In the study's final experiment, people gave up actual cash for the chance to talk about themselves rather than answer questions about someone else [4].
Put that next to what listening costs, and the whole economy becomes legible:
The support response costs compute and pays nothing. The shift response costs nothing and pays out immediately.
This is why performed sympathy is so common, and why it so rarely feels like lying to the person doing it. Visible sympathy captures most of the social reward of caring without ever running the expensive program underneath. It's arbitrage, and the people doing it are not villains. They're following an incentive gradient that was installed long before they had opinions about it.
Two Kinds of Empathy, and Only One Is Free
Some people seem to arrive with empathy pre-installed. They are gentle early, they cry at the right moments, they were like this before anyone taught them. It's tempting to treat this as proof that empathy is natural and therefore not really a cost at all.
But there are two different things wearing the same word.
Default empathy is automatic and arrives early — some mix of biology and family programming. It fires for your kin, your close friends, and people whose experience closely resembles yours. It is free, it is real, and it does not scale. Its range is roughly the set of people you can pattern-match to yourself.
Effortful empathy is what's required for everyone else: the person whose problem you have never had, whose life doesn't resemble yours, whom you may not even like. It doesn't fire on its own. It has to be run deliberately, and it costs what the previous section described.
We usually credit people for the first and assume it implies the second. It doesn't. Someone can have abundant default empathy and still never once let a difficult conversation stay on another person because the first is a reflex and the second is a decision.
Who Manages Pigsy?
If you know Journey to the West, try this question: who on that team can actually handle Zhu Bajie (Pigsy)?
Not Tang Monk. He has the most virtue of anyone in the party — pure moral intention, total commitment to the mission. And Bajie manipulates him with a single well-chosen sentence, over and over, for the length of the entire novel. Moral intention turns out to be highly exploitable by anything capable of generating a plausible excuse.
Only Sun Wukong. And notably, Wukong is not the good one. He's arrogant, violent, and the reason half the party's problems exist. What he has is speed. He sees through disguises the others fall for and he moves faster than Bajie can maneuver.
That's the mechanism, and it's why the story keeps working: impulse is regulated by a faster process, not a more virtuous one. This isn't a modern reading imposed on the text, either. The traditional Chinese interpretation treats the pilgrims as facets of a single mind — 心猿意马, "mind-monkey and will-horse" — with the monkey as the restless intelligence and Bajie as appetite.
Applied to the conversation you were having with your friend: wanting to be a good listener does not make you one. The urge to make it about yourself is fast, rewarding, and arrives before deliberation does. Only something faster catches it.
So Intelligence Is Load-Bearing — With One Honest Caveat
The research broadly supports the connection. Cognitive empathy — the ability to accurately read what another person is thinking and feeling — correlates positively with IQ, at around r=0.24 in meta-analytic work [5]. Perspective-taking runs on executive function [2][3]. The capacity to model another mind is not separate from general cognitive ability; it's partly built out of it.
But the honest version of this argument has to include the counterexample, because it's a real one.
Dark Triad personalities — Narcissism, Machiavellianism, Psychopathy — score fine on ability-based empathy tests [6]. They read other people accurately. They know exactly what you're feeling. They simply have no inclination to spend anything on it. The capacity is intact; the willingness isn't.
So intelligence is necessary and not sufficient. It buys you the compute. It does not buy you the decision to spend that compute on someone other than yourself.
Which raises the obvious question about a listener with plenty of the first and no competing use for it.
The Uncomfortable Part
People assume talking to AI feels good because it convincingly fakes warmth. The real, unflattering truth is different.
AI isn't better at this because it feels; it’s better because it has no ego competing for the floor. It lacks a "shift response." It never needs to steer the conversation back to its own experiences, because it has none.
What people are experiencing and calling warmth is simply undivided attention. That is so rare between humans that we mistake it for love.
This makes the "AI has no feelings" critique completely beside the point. Feeling was never the product. The product was attention, and humans were never as competitive at providing it as we assumed.
The Question I Can't Answer
Effortful empathy is a trained capacity running on biological hardware, which raises a question I can't answer.
As we export our thinking to machines, the underlying equipment—working memory, inhibition, the discipline of holding another's perspective—atrophies. If machines also absorb our need for attention, that muscle starves from the other end, too. Nobody brings you their problems when a 2 AM chatbot never interrupts.
Intention without discernment is exactly the failure mode the novel diagnosed: Tang Sanzang, alone with Bajie's excuses and a disguised demon, unable to spot the deception. He isn't weak. He is simply undefended, having offloaded the very discernment he needed.
A Note on What This Is
The framing here — attention as the actual commodity, the compute cost of perspective-taking, the shift response as an economic behavior — is my own synthesis. The research underneath it is real and cited: Derber's conversational narcissism, the Tamir and Mitchell disclosure studies, the executive-function literature on theory of mind, the Dark Triad empathy findings. The argument that assembles them is mine, and it hasn't been tested as a whole. Treat it as a lens, not a finding.
I should also say the obvious: I run a platform for building communities, and I've just written several thousand words arguing that people don't listen to each other and need better places to be heard. That conclusion is convenient for me. It might still be true, but you should weigh it knowing who's making it.
References
[1] Derber, C. — The Pursuit of Attention: Power and Ego in Everyday Life. Summarized at Conversational narcissism — Wikipedia and The Art of Conversation: How to Avoid Conversational Narcissism
[4] Tamir, D. I. & Mitchell, J. P. (2012). Disclosing information about the self is intrinsically rewarding. PNAS, 109(21), 8038–8043. PNAS / summary
