A well trained pilot would know when to trust the computers and when not to. They would also know how to maneuver and react in situations. It's like the pilot that landed his plane in the river after losing an engine to birds. I don't think a computer would have taken that option and not only would it have been likely that all the passengers would have been killed, but bystanders as the planes computer attempted to correct and eventually goes down in a populated street.
This comment looks sensible on the face of it, but I have to disagree with you. I have a pilot license and am familiar with the process of flying. I've never flown a fly-by-wire aircraft, but I've automated a radio broadcast desk - which might not look like it's relevant, but it taught me that "knowing when to trust the computer" is not an obvious state, not in a radio station and I seriously doubt in a cockpit.
For me the final "aha moment" came when the computer was attempting to tell me something useful, but because I was concentrating on a completely different aspect of interacting with it, I completely missed the information. In my case it caused a few seconds of dead air on a radio station, nothing life threatening, but not human obvious either.
The challenge is not "when to trust a computer and when not to" - the challenge is "how do you get the information that the computer is using to the human in such a way that they can manage that input stream in a timely fashion. Stick shakers are an example of making use of an extra input channel.
Accidents in planes are rarely just one thing going wrong, they generally are a whole string of things. A computer in the mix just exacerbates the issue.
