Why deep-learning AIs are so easy to fool Artificial-intelligence researchers are trying to fix the flaws of neural networks.