Neural networks are sometimes called black boxes because, despite the fact that they can outperform humans on certain tasks, even the researchers who design them often don’t understand how or why they ...