Because humans have invented tools - math, notably - that they can use to be more formal and precise when they deal with problems of this sort. To date, neural

Because humans have invented tools - math, notably - that they can use to be more formal and precise when they deal with problems of this sort. To date, neural networks can't -- they're doing very well at the kind of "intuitive" pattern matching that our brains do, but less well on an algorithmic front. It's one of the really interesting frontiers of AI right now. The folks at DeepMind recently showed, for example, that a variant of their "Neural Turing Machine" architecture could learn to do things involving more like what we think of as discrete algorithms (https://deepmind.com/blog/differentiable-neural-computers/ https://deepmind.com/blog/differentiable-neural-computers/ ), but all of this is very early work.