The usefulness of logic programming lies in constructing something that is provably optimal, but for practical applications such as self-driving cars and pattern recognition, suboptimal solutions such as gradient-based search and recurrent neural networks are more pragmatic.