fdr derjodel and others above
We have the same discussion after practically every accident and incident that involves loss of control, cfit, etc. The fundamental problem is that humans (some of them) are good at doing one thing, while automation (all of it) is good at something else. Automation, as long as it works, is good at performing precisely and flawlessly mundane repetitive tasks, precisely those where humans are likely to make mistakes because of boredom, complacency, tiredness, distraction. How many times in living memory did it happen that an autopilot failed to level off in ALT mode at precisely the preset altitude? Exacly none... How many times did it happen that those twitchy fingers turned that knob to the wrong altitude... ? On the other hand automation is incapable of performing anything outside the limits of what it was designed to do, including recognising its own failure. Humans on the other hand are quite capable of thinking creatively and out of the box, and can process and act upon a vast amount of visual information very rapidly, especially in situations they have been preconditioned to.
The fundamental problem is that humans are not equal (despite all the wishful thinking by some). Some have better skills than others in different areas. It is true that piloting an aircraft 50-60 years ago required above average skills in many areas. However the regular appearance of smoking holes around the world showed that even these above average individuals were prone to making mistakes, and sometimes simply the situation turned so complex that the human brain could not catch up with it within the time frame available. With the onset, development and improved reliability of automation, these situations have been vastly reduced, and piloting an aircraft no longer requires exceptional skills, anyone with a good average skill set can do it safely as millions of takeoffs and landings prove around the world. However the advance of automation brings about one aspect that is very difficult to deal with. It now requires way above average skills to recognise and anticipate the various situations that may arise if one or more components of the surrounding comfortable automation fails, and at least average or better skills to recover from an unanticipated situation. By definition, at least half will fail with the latter, but recognising those who will in advance is very difficult because training by nature conditions for anticipated situations.