To solve problems in science, engineering, and business, computers were first programmed with the explicit instructions to solve those problems. Now, AI has shown that it is more powerful to first train computers to learn and then give them the data needed to solve a problem. I will review the successes and limitations of machine learning methods being used to train: 1) quantum computers with only Dirac operator gates, 2) hybrid classical-quantum computers with variational quantum circuits, 3) quantum Hopfield computers using equilibrium propagation, a quantum replacement for back propagation, and 4) quantum computer annealers.