One of the big promises of open and massively online education is easy data collection: we can record everything from students’ habits in reading and viewing lectures, to their participation in discussion groups, to their timing and performance on exercises. So, open education is a natural fit for machine learning - for example, we can use ML to predict future student performance, to select and sequence learning activities, and even to help grade some types of assignments. But there’s a lot more left to do: I’ll argue that even-bigger gains can come from ML that’s focused on understanding educational content and how students learn it, and on communicating this understanding to human educators. To achieve such understanding and communication, we need to take advantage of ML techniques including representation learning, structured learning, and exploration / experimentation.