There's only one problem in computer vision, and it's not unique to the field at all: balancing sensitivity vs specificity. Modern applications of ML don't solve that.