The Air France crash happened when the autopilot bailed out on facing a situation it couldn't handle. (And while you can design an autopilot to handle the specific case of the pitot tubes freezing up, there's no better course of action for the general case of "sensor inputs that make no sense according to the autopilot design"). If you automate more that's going to make pilots worse not better at handling the rare situations the autopilot can't handle.
The pitot tubes unfroze after a few seconds, so the autopilot was perfectly capable of working again, but the pilots didn't re-engage it and kept trying manually, causing the end result...
My one question is "Who the hell decided that averaging stick inputs is a good idea??
