Inspired by the results of a game-playing neural network, mathematicians have been making unexpected advances on an age-old math problem.