[constantcrying]

This also relates to vision models. The existence of adversarial attacks (e.g. imperceptable changes in the image drastically changing the output) essentially demonstrate that the model has not reached the point at which the network "understands" the generalized concept it wants to disinguish.

[zone411]

The same argument could apply to humans. For example https://en.wikipedia.org/wiki/Change_blindness.

[cubefox]

That's something else. The OP was talking about small changes in pictures causing a very different classification.

[constantcrying]

Not really an example, there are many ways human vision is flawed and can be tricked, but none are on the level of these adversarial examples. There imperceptible differences between an image lead to a category error.

Human perception can be ambigous, but minimal changes never cause drastic category errors.