Basically this is a well-known issue in aviation. As automation gets better and better, humans rely on it more and more and get less and less able to handle even minute failures in it. Additionally humans are absolutely terrible at being able to be dropped into a complex situation and have to make immediate rational decisions about it - thus the failure mode of "in case of error, tell human to handle it" is a bad idea if time between "tell human about it" and "crash into things" is under ten seconds or so. This has been the subject of many NASA studies and NTSB reports and the above article does a good job presenting this info in a form a layman can understand.
There are currently no known easy solutions, sadly.