If you look at something like chess, which was often thought of as very difficult, the first thing we designed that was really good at the game made us realize that you needed next to no logic to win if you had a computer fast enough to try a billion moves. But we thought object recognition would be super easy, far less than AI tier, and we still can't do it reasonably well.

I'm much more convinced that humans suck at estimating the hardness of problems, and not very convinced we keep cheating the classification of AI.