One of the limitations of any kind of automation is that it can only react within the parameters in which it was designed. That's why many systems will throw an alarm or go into alternate law, "conditions have just exceeded my ability to control them...your airplane."
As soon as the first "unable, say intentions" comes across, automation quickly reaches its limit.
The AF flight went into the ocean when the automation couldn't understand the inputs. Turns out the humans in that case weren't any better. Automation would have failed completely in Sioux City because it could not have been designed to have reacted to a situation so far outside anything a designer would have considered. The flight crew adapted, improvised, and overcame. We have a very long way to go before automation can do that.
Airliner, and GA, automation has come along at an incredible pace, Moore's Law is working. In cases now, an airline flight crew may hand fly for only a few minutes per flight. Maybe, in the near future, the aircraft will taxi, takeoff, cruise and land unassisted - under human command and supervision. But a human will still have to be part of the loop, even if it's just for the SHF moments.