Neural networks are not programmed like a conventional computer system, they are trained. That involves feeding the system with millions of example problems, and gradually tuning its connections until it produces good solutions most of the time. Finding, or creating, enough data to train the model is a key challenge in AI development.