You do it dozens of times a day, and you almost never notice.
You glance at a photo and something in you says that's beautiful — before you could tell anyone why. You read a sentence and feel, in an instant, that it's slightly off, a word out of place you haven't consciously identified yet. You walk into a room and sense it's almost right, that one thing could be moved and it would come together. You look at a design, a melody, a face, a line of code, and a judgment arrives — fast, certain, and ahead of any reasoning you could give for it.
We treat this as nothing. It's just taste, just an opinion, just a feeling. But it is one of the strangest and most quietly remarkable things a human being does — this instant, pre-rational knowing of what is good, what is wrong, what could be more. And it is precisely the thing that AI, for all its astonishing fluency, does not seem to have.
I want to sit with that, honestly, because I think the usual way we argue about it is wrong. The question everyone asks is "can AI be creative — can it make art, music, writing?" And the answer, increasingly, is yes, it can produce those things, impressively. But I don't think production was ever the point. I think the human thing — the thing that might be genuinely out of reach — is something else entirely. Not the making. The discerning.
The distinction almost everyone misses
Here is the move that the whole question turns on, and most of the debate skips right over it.
We lump two very different capacities together under the word "creativity," and they are not the same thing at all:
Generation is producing output — a poem, an image, a tune, a layout. And I'll say this plainly, because pretending otherwise would make me a liar and cost me your trust: AI is extraordinary at generation now. It can write a competent poem, render a striking image, compose a passable melody, draft a clean design. If your claim is "AI can't produce aesthetically pleasing things," you have already lost the argument, because everyone has seen that it can.
Discernment is judging that output — knowing which of two versions is better, sensing what's wrong, perceiving what could be improved, having taste. And this is a completely different faculty.
Here's why the distinction matters so much: generation was never the hard part of creativity. A random number generator generates. Throw enough paint at enough walls and you'll produce something. The miracle of the creative human was never that they could make marks — it was that they could look at the marks and know. Know which one was alive and which was dead. Know what to keep and what to throw away. Know, standing in front of the thing, that it wasn't finished yet, and feel in which direction "finished" lay.
The discerning mind, not the producing hand. We got so dazzled — or so threatened — by watching machines produce that we forgot the production was never where the magic lived.
What discernment actually is
So let's look closely at this strange faculty, because the closer you look, the stranger it gets.
Discernment isn't a rule you apply. You don't compute that a face is beautiful or a phrase is clumsy. You perceive it, the way you perceive a color — directly, immediately, often before a single word of justification is available to you. The reasons come after, if they come at all, and half the time they're reconstructions, stories we tell to explain a verdict we'd already reached.
And it's deeply embodied and personal. Your sense of what's beautiful is tangled up in a lifetime of sensory experience, of things you've seen and heard and touched, of memories, of emotions, of the particular person you happen to be. Two people with identical knowledge can have completely different taste, because taste isn't knowledge. It's a point of view. It's yours.
But here is the part I keep coming back to, the part that I think matters most: discernment comes from caring. You can only find something beautiful if things are capable of mattering to you. The whole faculty is downstream of being a creature that experiences, that prefers, that is moved, that has something at stake in how the world is. You judge beauty because beauty does something to you. Take away the one who can be moved, and I'm not sure "beautiful" means anything at all — it's just a word with no one behind it to feel its weight.
Which brings us to the machine.
Why AI may lack it, structurally
An AI that generates aesthetic output is doing something genuinely impressive, but notice what it's doing. It has learned, from an ocean of human-made examples, the statistical shape of what people have called good. When it produces something beautiful, it is, in effect, finding the center of mass of human aesthetic preference and generating toward it.
And here's the problem hiding in plain sight: real taste is almost the opposite of an average.
Taste is particular. It's a point of view, which by definition is not everyone's. The most interesting aesthetic judgments in history were often the ones that cut against the consensus — the thing nobody had learned to love yet, that one person saw and insisted was beautiful before the rest of us caught up. An average can tell you what people have found beautiful. It cannot want something to be beautiful in a way no one has wanted before, because wanting requires a wanter, and an average is nobody's.
But the deeper issue isn't statistical, it's this: there is no one home. The model has processed every description of awe ever written, every account of grief and longing and wonder and beauty — and it has never once been moved by anything. It knows, with superhuman thoroughness, what beauty is described as. It has never felt it. Not once. Not for a moment.
And so when it "judges" — when it tells you which image is better, or improves your paragraph — I think something is missing that we're too quick to overlook. A judgment with no subject behind it, no one who is moved by it, nothing at stake for anyone in the judging — is that judgment, or is it an extraordinarily good imitation of the residue that judgment leaves behind? The machine has the map of human taste, in staggering detail. But a map with no one reading it, no one caring where it leads, is just ink. The discerning mind might be the reader, not the map — and that's the part I'm not sure can be manufactured.
Now let me argue against myself
If I stopped here, I'd be doing the thing I dislike — telling you a comforting story about human specialness and calling it an argument. So let me give the other side its real strength, because it has some, and if you've been thinking skeptically this whole time, you're right to.
Maybe what we call "taste" is just extremely sophisticated pattern-matching on human preference — and the feeling that it's something more, something sacred, is simply human exceptionalism. We have an embarrassing track record here. We said machines could never play chess, then never play it creatively, then never master Go's intuition, then never use language — and each time, the thing we'd called uniquely human turned out to be a pattern a machine could learn. "Aesthetic judgment is special and unreachable" might just be the next item on that list, and I might be writing the same confident sentence people wrote right before they were proven wrong.
And maybe — this is the uncomfortable one — the "experiencing subject" I keep invoking is itself not what I think it is. Maybe we are also prediction machines, evolved to pattern-match on our environment, and the thing we call "being moved" is just what our pattern-matching feels like from the inside. If that's true, there's no magic line here at all. The machine does with silicon what we do with meat, and my insistence that ours has a someone behind it and theirs doesn't is just a bias I have because I happen to be one of the meat ones.
And purely practically: maybe AI doesn't need to feel beauty in order to reliably identify what humans will find beautiful — and if it can predict your aesthetic response perfectly, then for every practical purpose that touches your life, the difference between "real taste" and "flawless simulation of taste" may be a difference that makes no difference.
These are not weak arguments. I don't have a knockdown reply to any of them. Anyone who does is selling you something.
Where I land, held loosely
So, with all of that honestly on the table — here is what I still believe, offered as a suspicion and not a proof.
I think discernment is distinctly human not because the machine isn't powerful enough, but because aesthetic judgment may be inseparable from being a subject with stakes. Taste, real taste, isn't only identifying what people like. It's caring that something is beautiful. It's being changed by it. It's the whole circuit — being moved by something, and out of that being-moved, wanting to make something that moves someone else the way you were moved. That circuit runs, at every step, through an experiencing self. And a machine can reproduce the outputs of that circuit — the beautiful image, the confident judgment — without any of the circuit actually existing.
A judgment that nothing is at stake in, that no one is moved by, that doesn't matter to anyone doing it — I keep arriving at the same place: I don't think that's quite judgment. I think it's a very good imitation of what judgment leaves behind.
So will AI ever reach the discerning mind? My honest answer is: I suspect not — but not because it can't become more capable. Because the thing we're pointing at might not be a capability at all. It might be a property of being someone. You can't scale your way to a subject. And I hold even this loosely, because the truth is nobody knows, and the history of "machines will never" should make all of us humble, in both directions.
What we're left with
Whatever AI becomes, the discerning mind is, for now, still ours — that small daily miracle of looking at something and knowing, of being moved before you can explain why, of sensing that a thing could be more and feeling, dimly, in which direction.
It's easy to feel diminished watching machines generate in seconds what used to take us years. But maybe the quiet consolation in all of this is that generation was never where our worth lived. The seeing was. The being-moved was. The this could be better was. Those might turn out to be the last things to go — or they might be the things that were never really on the table at all, because they were never about output in the first place. They were about being alive to the world.
The machine makes. We discern. And discernment, when you follow it all the way down, might just be another name for caring that things are one way rather than another — for the world mattering to someone.
That's not a small thing to be left holding. It might be the whole thing.
Honest question, because I genuinely go back and forth and I'd rather hear real thinking than applause: is taste a skill that enough data could eventually teach — or is it inseparable from being a conscious, caring subject? And here's the one that keeps me up: if a machine could perfectly predict what you'd find beautiful, while feeling nothing at all — would that be taste, or a flawless imitation of it? And is there even a difference? I don't think I know.
Top comments (0)