Deep learning often feels like an empirical science still waiting for its theory. I’m interested in what the conceptual upheavals of early quantum mechanics suggest about how we should interpret (and trust) the high-dimensional representations our models learn.