The question of how machines can be endowed with the ability to acquire and robustly manipulate commonsense knowledge is a fundamental scientific problem. Here we formulate an approach to this problem that we call knowledge infusion. We argue that robust logic offers an appropriate semantics for this endeavor because it supports provably efficient algorithms for a basic set of necessary learning and reasoning tasks. We observe that multiple concepts can be learned simultaneously from a common data set in a data efficient manner. We also point out that the preparation of appropriate teaching materials for training systems constructed according to these principles raises new challenges.