“People believe the best way to learn from the data is to have a hypothesis and then go check it, but the data is so complex that someone who is working with a data set will not know the most significant things to ask. That's a huge problem.”