Machine learning systems can improve at tasks through experience, but improvement alone does not settle whether they understand what they have learned. The answer depends on how learning is defined—and that distinction matters when evaluating AI capabilities.