AI taught to beat Sudoku puzzles. Now how about a time machine to 2005?

AI can now solve some of the hardest Sudoku puzzles to a high degree of accuracy, according to new research that teaches machines to logically reason. Sudoku was, if you can recall, all the rage in the West at least a decade ago. There are several techniques and algorithms to crack the puzzles within a second, however, it's an interesting problem for deep-learning software to tackle, as it's an exercise for neural networks to practice complex reasoning. A paper describing the method that uses recurrent relational networks was published through ArXiv earlier this month. It builds upon DeepMind's previous work with relational networks (RNs), a type of neural network that focuses on the relationship between pairs of objects.