The "black-box" nature of deep learning may hinder trust in critical applications, such as legal or medical decisions.