You should be aware that echo-state networks are considered special cases of recurrent networks. One variant of recurrent networks, LTSMs, dominates state of the art on a wide variety of problems.

They're expensive to train, in computing time and data-requirements, but they're the closest thing we have to trained programs.