Ravi

That direction of error is the dangerous one, because it hands anybody a way to dismiss a real photograph. Easier argument with students: ask them to find who published it first. If they cannot, the picture is unusable regardless of what any score says.

The ordering is the part worth defending. People reach for the detector first because it is one click and it returns a number, and the number is the least reliable thing in the whole process.

On a picture desk the first question is always who had it first, and it resolves most of them before anybody looks at pixels. The reason it works is not clever: a photograph of an event exists because a person was at the event, and that person is findable. A generated image has no such person, and the absence shows up as a trail that stops.