Look closely at any machine-learning algorithm and you’ll inevitably find people—people making choices about which data to gather and how to weigh it, choices about design and target variables.