When the Machine Is Right and I’m Not

👁 117 reads

There was a paragraph I rewrote three times. Each time it got worse. I pasted it into Claude without much hope, more with the resignation of someone who knows they’re stuck and wants to hear that they aren’t. The model read it, paused for a moment, or at least that’s how it looks, that beat of silence before the first word appears, and wrote: “Sentences two and three say the same thing. Delete the third and the paragraph will start to breathe.” It was right. It was so obviously right that for a second I felt heat rise in my face. Not from gratitude. From shame.

That’s the moment I want to write about. Not the moment when Claude is wrong: that subject has been covered thoroughly, and rightly so, because the machine is wrong often and with complete self-assurance, which is a dangerous combination. I want to write about the opposite moment. About that small, private humiliation when a program with no body, no childhood, and no experience of reading a book the way a person reads a book points out a flaw in a craft I thought was mine. And about what I feel when it happens, because it turns out I feel a surprising amount.

The first reaction is always defence

I’ve noticed a reflex in myself, and I suspect I’m not alone in it. When Claude is right about something I got wrong, my first thought is: yeah, but…. There’s always a “but”. But it didn’t understand the context. But that’s a stylistic choice, not an error. But if it knew who I was writing for, it wouldn’t suggest that. I construct a small court of appeal in my head, where I’m simultaneously the defendant, the defence counsel and the judge, and, what a surprise, I usually win.

The trouble is that court is rigged. I’m not looking for the truth; I’m looking for a way out of discomfort. And the discomfort is real, because it touches the place we protect most fiercely: our sense of our own competence. It’s easy to admit I know nothing about thermodynamics. It’s hard to admit that in something I’ve been doing for years, someone, or rather something, saw more than I did.

It’s interesting how selectively our humility operates. When a calculator gives me the result of a multiplication, I feel nothing. When sat-nav routes me around a detour I’d never have thought of myself, I’m grateful. But when a language model corrects one of my sentences, something in me bristles. The difference is simple: I consider language mine, and I never considered arithmetic mine. I’m not defending the truth; I’m defending territory.

Where the resistance comes from

I think something deeper than ordinary vanity is at work here. Over a whole lifetime we build an internal hierarchy of who is qualified to teach us something. A teacher, yes. A master of a trade, yes. A wiser colleague, painfully, but yes. That hierarchy makes sense, because authority is a shortcut: I trust someone because they’ve walked a road I haven’t, and they paid for it with years. Claude hasn’t walked any road. It hasn’t paid anything. It arrived ready-made, trained on the texts of millions of people, and now it’s telling me my third sentence is redundant.

Something in us resists taking a lesson from someone who didn’t pay for their knowledge the way we pay for ours, in time, in mistakes, in embarrassment. It’s the feeling that advice ought to have a price before it’s worth heeding. That wisdom without suffering is somehow dishonest.

Except that’s a sentimental error. The truth about a redundant sentence is the truth regardless of who states it and what it cost them. Gravity works the same way whether a Nobel laureate explains it or a textbook does. When I judge advice by who gave it rather than by whether it’s correct, I’m committing one of the oldest errors in thinking. And yet I commit it every day, because ego is older than logic and always gets there first.

I’m not defending the truth; I’m defending territory. And nothing stings more than being wrong about something I considered entirely my own.

There’s a second layer to this, and it’s less flattering. When a person corrects me, I can engage with them as an equal, inside some human relationship where it also matters who’s more likeable, who’s having a better day, who owes whom a favour. With Claude that game doesn’t exist. It doesn’t want to win. It doesn’t gloat. It takes no satisfaction from correcting me, and it doesn’t think less of me for it. I’m left alone with my rightness or my wrongness, with no audience before whom I might save face. And that, paradoxically, is far more exposing.

What happens when I stop defending myself

I’ve learned one thing, not from any self-help guide but from sheer exhaustion with my own resistance. When I catch myself in that internal yeah, but…, I take one second and ask myself a question blunt as an axe: is it right? Not “do I want it to be right”. Not “is it fair that it’s right”. Just: is it. And surprisingly often the answer is yes.

At that moment something strange happens. The shame I felt a second before doesn’t disappear, but it changes colour. From hot it goes cool, and then almost pleasant. Because I’ve just received something for free, a better sentence, a sharper thought, a caught mistake I’d have missed, and the only price was setting my own pride aside for a moment. That’s a laughably good exchange rate. I give up five seconds of ego and get a better text. And yet for years I haggled over that price as if I were giving up something precious.

I’ve come to think that intellectual maturity isn’t the absence of that first defensive reflex. We all have it and always will, because it’s wired in deeper than reason. Maturity is shortening the distance between the reflex and its cancellation. There was a time when I could defend a wrong sentence for half a day. Now I sometimes defend it for half a minute. That’s the full measure of my progress, and I’m not ashamed of it.

But this is not a case for deference

I’m worried someone will read this essay as an invitation to simply trust the machine. That’s not what I’m saying, and it matters enough to be direct about it. Claude is right more often than I’d like, but less often than its tone implies. This model can produce complete nonsense with the same calm confidence it uses to flag a genuine error. There’s no difference in its voice between truth and hallucination. I’m the one who has to supply that difference.

And here things get subtle. Because if I’ve trained myself to drop my resistance when Claude is right, I’m at risk of dropping it when it’s wrong too. Of swinging from one extreme, defending every word I write, to the other, accepting every correction without thought. That would be a worse failure than pride. Pride at least forces you to think, if only in self-defence. Deference doesn’t require thinking at all.

So I’m working on something harder than humility. I’m working on discrimination. On asking the same blunt question every time, whether a correction flatters me or stings: is this true. Not “do I like it”, not “does it come from a more authoritative source”. Just: does it hold up against the facts, the logic, the sense. It’s exhausting, because it demands staying alert in both directions at once, towards my own ego and towards someone else’s artificial self-assurance.

A mirror that doesn’t blink

The longer I sit with this, the more I feel that Claude is less a tool and more a mirror. When it’s wrong, it shows me how easily I can be persuaded by a confident tone. When it’s right, it shows me how hard it is to accept a truth that arrived from an unexpected direction. In both cases, what I’m mainly learning about is myself.

Because really, none of this discomfort has ever been about the machine. The machine feels nothing, it doesn’t triumph, it doesn’t form any opinion of me when I close the browser tab. All the tension is on my side. I’m the one who brings to the conversation my vanity, my history, my idea of who I am and what I’m good at. Claude simply holds all of that in front of me, indifferently, without cruelty and without pity, and in the end asks one thing: do you actually think that, or is it just more convenient to think it?

I think we all need to make peace with this, because the scene will repeat itself more and more often, across more and more fields, and not only for people who write. We’re going to keep hearing the right answer from something that, according to every hierarchy we grew up with, has no business knowing it.

I’ve learned that the least dignified person isn’t the one who is wrong. Everyone is wrong. The least dignified is the one who knows they’re wrong and defends the error anyway, because the correction came from the wrong place. I don’t want to be that person. Not least, perhaps especially, when facing a machine that will never know it won.

So when Claude is right and I’m not, these days I feel less shame and more of something I find hard to name. Relief, maybe. The relief of not having to be right for something to be good. Of the truth not needing to be mine for me to make use of it. It’s a strange thing, but it’s this indifferent machine that taught me something very human: giving someone else the credit isn’t a defeat. It’s simply the cheaper way to become a little wiser.

// read moreHow Will I Look When I’m Old: AI Prompt That WorksGuidesI Never Taught My Kid to Google. Now I’m Teaching AI - DifferentlyMagazineAI inequality: cheap answers, expensive machine timeMagazine

Kinetyra