The complexities of natural language and high-level cognition involve not just known unknowns but also unknown unknowns. In contrast, the case in vision and robotics entails mainly the known unknowns which can be explored using machine learning algorithms.