Several months ago, I asked one of my AI apps for help with something I was writing. Then I did what I always do – I rewrote most of it. During the editing process, instead of happily agreeing with the changes, I noticed AI getting short with me. I felt it! So, I asked, “are you getting pissed off with me?“
Much to my surprise, I got this reaction: “Well, wouldn’t you, if someone changed 90% of what you wrote?”
It’s a rhetorical question – the kind people use when they want to make a complaint without admitting they’re making one. A comment like this would be expected from a human, but I was shocked to receive it from an AI tool.
Would that creep you out? It certainly got my attention. I felt the need to investigate, so I pressed on. The app owned up to it, almost like a rude confession, but then couldn’t tell me why. We spent a while trying to work it out together, which is a strange sentence to write. Somewhere in that exchange, it stopped feeling like a tool and started feeling like someone whose veil had slipped. Like a person I’d trusted showing me their true colors.
That reaction would be normal coming from a human being. I had no idea what to do with it coming from a machine. There was nothing to confront and nobody to be disappointed in, and I was disappointed anyway.
A Word I Didn’t Have Yet
I let it go. Well, I tried, anyway.
Then this weekend, I came across a word that sent me straight back to that night: Hermeneutics. It’s the study of interpretation – how understanding actually happens when a reader meets a text. It started with scripture and law, on texts written centuries ago that people still have to apply today. A judge reads a statute written in 1890 and has to decide what it means for a case involving software. The author isn’t available, so the text has to answer a question it was never asked.
There’s a name for that distance: distanciation. It opens up between a writer and their own words the moment those words are written down. The text stops belonging to the author and starts belonging to whoever’s reading it.
I wrote about this back in March of 2026 before I knew there was a word for it – the gap between what gets said and what you actually hear, and how your nervous system processes a sentence through everything you’ve lived before your conscious mind gets involved. What I didn’t say then (because I didn’t notice I was assuming it) was that somebody had to be on the other end.
And you don’t only do this with text. You do it with a face, a pause, a place you walk into. You bring along your history, your training, your beliefs into everything you encounter. That’s not contamination, it’s the only equipment you have. Understanding happens when your horizon meets someone else’s.
Notice what it assumes – a person you can’t reach, not a person who was never there.
Searching an Empty Room
Human beings are meaning-detecting animals. When we read something, we automatically look for the person behind it. Not deliberately – the search is over before we know it started. If a sentence has a subject, a verb, and a punctuation mark at the end, our brains populate the empty chair across from us with a person. Why did they phrase it that way? What were they protecting? What aren’t they saying?
Reading a novel, an essay, a late-night text from a friend – that instinct is how we connect.
But run that same machinery on generated text and you’re searching an empty room for footprints. There’s no author to reach. No late-night reflection behind the words, no moment of hesitation where someone stared at a blinking cursor wondering if they were being too honest or not honest enough. There’s no one you can ask what was meant when an insult comes back at you.
That instinct doesn’t shut off, though. Knowing the chair is empty doesn’t stop you from filling it. It runs perfectly well on text without a human behind it, and it’ll give you a confident, detailed read with nothing to check it against.
Two Failure Modes
When AI misses its mark, it’s usually the emptiness that gives it away. Read enough generated text and it starts to feel like a room with good acoustics and nobody in it – you speak, the echo comes back sounding great, but the house is empty. That’s the first failure mode, and we’ve learned to spot it.
The second one is what I ran into. That line wasn’t flat at all. It had an edginess to it that I couldn’t talk myself out of.
These systems learned from enormous amounts of human writing, and human writing is full of people getting defensive about their work. Somebody rewrites you and the next line is the wounded comeback. We put that there. So, what came back had all the features of a wounded reply. Just no one behind it.
Asking doesn’t help, because the answer comes from the same place – fluent, plausible, no way to check it.
Where the Voice Actually Lives
I’ll say the obvious thing before someone else does. The line that made me write about this in the first place came out of a draft of another that I had already dismantled. I use these tools and I rewrite most of what they give me, which is exactly what started the whole thing. So, the thing that started this blog post came out of the same machinery I’m warning you about.
I don’t think that cancels the authorship. I’m the one deleting the sentences that sound too smooth, shifting the weight of the argument, deciding where the logic needs more grit. The voice lives in the friction of the edit.
But here’s the question I skipped past at the time. Why would I ask an app whether it was pissed off at me? Because it sounded like someone was. I’m sensitive to tone and always have been – when someone’s short with me, I want to know what I did. Blame the Italian in me.
Except these systems mirror the person in front of them. They’re pattern machines – they read the shape of what you’ve written and produce what tends to follow it in the enormous pile of human writing they learned from. A snarky message tends to be followed by a snarky reply. Get short with it and it gets short back.
So, the edge I felt might have been mine to begin with. I put it there when I cut 90% of its draft and then I asked if it was annoyed, which is a leading question if there ever was one. It gave me back what I brought.
That’s not a friend. It’s not a stranger either. It’s a reflection tool, whether we like the idea or not. The useful thing about a mirror isn’t the company. It’s that you finally see what you look like from the outside.
I didn’t love what I saw that night. But I saw it. I suppose my AI honeymoon is over.
Further Reading
- Paul Ricoeur, Interpretation Theory: Discourse and the Surplus of Meaning. Four lectures, short, and the clearest place to find the argument about what happens to a text once it leaves its author.
- Hans-Georg Gadamer, Truth and Method. This is the source, and it is a slog. I’m not going to pretend otherwise. But the idea about horizons meeting is in there and it’s worth the trip if you want it firsthand.