Theoretical physicists from ETH Zurich deliberately misled intelligent machines, and thus refined the process of machine learning. They created a new method that allows computers to categorize data—even when humans have no idea what this categorization might look like.