I think "real AI" could help (because it could also explain as well as configure). Systems that can't explain themselves (and most can't) are a very bad idea.
Either way, the idea of machines or systems as "living" and able to communicate intent and process, even if only within their own "umwelt", is really interesting. Even a taste of that would make modern systems easier to debug and understand, if not more robust (which would be a better starting point for many systems anyway I suppose).
Neural networks definitely are black boxes, at least at an individual level. Sure the concept remains the same generally, but the internals are different and hidden from case to case