We care about the existence of a nontrivial set of images that demonstrate a troubling lack of robustness in image classifiers, at least until we we have good reason to say with confidence that such failures will not be a problem in practice.