One problem is that "statistical inference" is overly reductive. Sure, there's a statistical aspect to the computations in a neural network, but there's more to it than that. As there is in the human brain.