The photograph that was not taken

In 2023 a picture of a flooded street in a European city spread across social networks during a storm. It had everything a news photograph needs: water to the windows of parked cars, a bicycle on its side, grey light. It had been made by software, and it showed a street that does not exist.

What is interesting is not that it fooled people. It is why it fooled them. The picture matched what readers already expected a flood to look like, because the software had learned from thousands of real flood photographs. It was, in a sense, the average of every flood picture ever published. Nothing in it was surprising, and nothing surprising is exactly what makes an image feel ordinary and therefore true.

Real photographs are rarely average. They contain things the photographer did not intend: a shop sign half out of frame, somebody looking the wrong way, a reflection that does not quite make sense. Those accidents are a kind of evidence, and generated pictures tend to lack them because nothing in them is accidental.

This gives readers something practical to look for, at least for now. Not fingers and shadows, which the software is steadily getting better at, but specifics. Where exactly is this? What is the name on the shop? Who took it, and what else did they take that day? A real photograph is attached to a place and a person. A generated one is attached to nothing, and when you ask it where it came from, there is no answer.