On a recent podcast, a host compared AI-generated content to a magician sawing a woman in half. You hear this a lot: AI outputs are illusion, something that looks real but isn’t.
That’s a category error.
When a magician “cuts” someone in half, no one is actually cut. Reality and appearance diverge. But when a model generates Python code, it is real Python code. It runs or it doesn’t. There’s no hidden assistant holding the application together.
You can expose the magician’s trick by examining the woman more closely. But examining code will not allow you to prove it was written by a human.
Digital artefacts are numbers, not atoms. Matter has physical identity. Numbers don’t. Digital copies are indistinguishable from originals, and the thing itself contains no trace of its origin. So arguments about “AI authenticity” aren’t claims about the output, they’re claims about provenance.
That’s a very different problem.
Does provenance matter? I hear people say they feel cheated discovering a song they loved was AI-generated. They feel that art has less value without a human story behind it. You might also be concerned that you’re being charged for more hours than the task actually took due to automation.
That’s legitimate. But it means authenticity can’t be solved by slapping a “Created by AI” label on things. That approach is operationally burdensome, trivially easy to evade, and focused on policing outputs rather than clarifying expectations.
Once you understand if your claim is about the bits or the provenance, you can focus your attention in the right place.
Leave a Reply