[ben_w]

Recent update — we do know, and humans are vulnerable to perturbations created to fool a multitude of existing AI: https://arxiv.org/pdf/1802.08195.pdf