What AIs are not Learning (and Why)

It is hard to make robots (including telerobots) that are useful, and harder still to make autonomous and collaborative robots that are robust and general. Current smart robots are created using manual programming, mathematical models, planning frameworks, and reinforcement learning. These methods do not lead to the leaps in performance and generality seen with deep learning, generative AI, and foundation models (FMs). Today's robots do not learn to provide home care, to be nursing assistants, or to do household chores nearly as well as people do. Addressing the aspirational goals of service robots requires improving how they are created. The high cost of bipedal multi-sensory robots ("bodies") is a significant obstacle for both research and deployment. A deeper issue is that mainstream FMs ("minds") are not created by agents sensing, acting, and learning in context in the real world. They do not lead to robots that communicate well or collaborate. They do not lead to robots that learn by experimenting, by asking others, and by imitation learning as appropriate. In short, they do not lead to robots that are ready to be deployed widely in human service applications. This paper focuses on what human-compatible service robots need to know. It recommends developing "robotic" FMs based on diverse experiential data.