Generative adversarial networks are not just good for causing mischief. They can also show us how AI algorithms “think.”
