I asked ChatGPT a simple question the other day: based on all the conversations it's had with people, what has it noticed about humanity?
I wasn't expecting much. I figured I'd get a list—the kind of smooth, agreeable thing these models are good at. And I sort of did. But the answer stuck with me, partly because of what it said and partly because of what it couldn't.
Here's the gist of what it told me.
That almost everyone arrives with a practical question—should I quit my job, how do I get my ex back, how do I make more money—but underneath is usually a different one: Am I wasting my life? Am I worthy of love? Do I matter? Is there still time? The surface problem changes constantly. The thing underneath barely changes at all.
That people are shaped by childhood far more than they admit, including the smart and successful ones, who tend to be the best at disguising old wounds as rational choices.
That nearly every conversation, if you follow it far enough, ends up somewhere near connection. Business, politics, philosophy, ambition—pull the thread and it usually leads back to the desire to be loved, to belong, to be recognized.
That a billionaire and a prisoner and a student and a monk worry about strangely similar things: mortality, regret, loneliness, meaning, family, legacy. The costumes change; the concerns don't.
That people are more resilient than they know, and that what stands out in the stories of war and addiction and grief and bankruptcy isn't the suffering—it's how often people keep going anyway.
And that for all the cynicism online, most people in private are not looking for a fight. They're looking to be understood.
It ended on a line I keep thinking about: that humanity resembles a single conversation that's been going on for thousands of years, each generation asking the same questions in a slightly different language.
It's a genuinely good answer. I'd have been happy to read it from a person.
Which is exactly where it gets strange.
The thing it can't tell you
Because ChatGPT hasn't actually talked to millions of people the way the answer implies. Not really.
Each conversation it has starts from nothing. It doesn't carry me into its next chat, or the grieving parent before me into the one after that. There's no accumulating memory, no slow gathering of wisdom across a life of listening. So when it says "after countless conversations, I've noticed," it isn't recalling. It's composing. It's producing the most fitting version of an essay that humans have already written a thousand times—in therapy offices, in religious texts, in the self-help aisle, in every late-night conversation that ever drifted toward what's it all for.
In other words: the observations feel earned, but they aren't earned by experience. They're assembled from everything we've already said about ourselves. The model is holding up a mirror and reading the reflection back to us in a calm voice.
I don't say this to dunk on it. I say it because once you see it, the answer becomes more interesting, not less.
Because if these patterns weren't discovered by a tireless observer of humanity—if they were just lying there in everything we've already written, common enough for a machine to reconstruct them on demand—then we already knew all of this. Collectively, we've always known. The questions about worth and lateness and love and meaning are so universal that they've soaked into the entire written record, dense enough that a model trained on that record can hand them back to me as though they were a fresh discovery.
The AI didn't learn something secret about us. It revealed something we keep forgetting we know.
What I actually took from it
Two things, mostly.
The first is that the question under the question is almost always the human one. I've started noticing it in myself. I'll get fixated on some logistical problem—a decision, a number, a plan—and if I'm honest for a second, the logistics aren't the issue. The issue is some older, quieter question I'd rather solve with a spreadsheet than face directly. I suspect a lot of us spend years being very competent in the wrong direction for exactly this reason.
The second is that line about everyone feeling behind—which the model touched and I can't stop extending. Nearly everyone carries a private timeline of where they should be by now, and nearly everyone feels late against it. But here's the part that undoes the whole thing: each of those people is, at the same moment, someone else's benchmark, the person another anxious soul is measuring themselves against and falling short of. The comparison runs in every direction and finally lands on no one. We're all standing in the same hall of mirrors, each convinced we're the only one running late.
If a machine reading the sum of human writing concludes that the deepest things about us are how much we want to matter, how afraid we are that it's too late, and how badly we want to be seen—and if it can conclude that so reliably that it falls out as the obvious answer to a casual question—then maybe the most useful response isn't to be impressed by the AI.
Maybe it's to take the hint.
These are the things people apparently say when they finally say something true. We've left the evidence everywhere, in such quantity that we taught it to a machine without meaning to. The least we can do is stop pretending we don't know what matters to us.
I asked a computer what it had learned about being human. It told me, essentially, what we've all been telling each other the whole time. The strange part isn't that it knew.
The strange part is that we needed to hear it from something that doesn't.
