Modern machine-learning models, such as neural networks, are often referred to as “black boxes” because they are so complex that even the researchers who design them can’t fully understand how they ...