Yes, i'm aware of differential attacks on neural networks. That doesn't falsify the hypothesis. There are instances where you will fail to recognize a human that those NNs will not.