ai
Compared to Whom? The Hallucinations We Call "Conversation"
We love to criticize AI for hallucinating, but a closer look at human conversation reveals we do the exact same thing. Who is truly holding the higher standard?
The Dinner Party Truth
Everyone calls AI unreliable because it hallucinates. So do you. You did it last week and called it “making conversation.”
A while ago, I spent a whole evening explaining to friends why High German sits so close to the dialect spoken around Hanover. I had the story completely ready: a Kurfürst von Sachsen, a castle conveniently near Hanover, delivered with the confident shape of a person who has read a few things. Nobody asked for a source. We had another round of drinks and moved on.
The Anatomy of a Human Hallucination
Writing this post, I went back to check that story, and most of it quietly fell apart. There really was a Kurfürst von Sachsen, Friedrich the Wise, whose chancery German (carried by Luther’s Bible and the printing press) became the backbone of modern Hochdeutsch. But the castle near Hanover? I completely invented that. The chancery actually sat in Wittenberg and up on the Wartburg in Thuringia, both well over 200 kilometers away. I had taken one real fact and welded a plausible bridge onto it—and the bridge was the part I made up.
What caught it was not a hacker or a historian. It was an AI. The exact same kind of model people enjoy calling unreliable. The difference? I had told it to back every claim with a source and never invent a name, a law, or a date. So it didn’t. It showed its work, which is more than I managed at that dinner.
Holding the Mirror Up to the Machine
That is the quiet hypocrisy of our current relationship with technology. We hold the machine to a standard of citation we never apply to ourselves, then act surprised when it clears the bar we keep tripping over. It is more reliable. Not perfect, but more.
So before we dismiss the model as unreliable... we have to ask ourselves: compared to whom?