The fact is that these adversarial examples are incredibly rare. 1 in a trillion trillion or more chance that they would occur in natural data. We dont know tha

The fact is that these adversarial examples are incredibly rare. 1 in a trillion trillion or more chance that they would occur in natural data. We dont know that human brains aren't vulnerable to something similar. I bet if we could back propagate through the visual cortex, we would find similar things. They can also be mitigated a bit by training on them.