Well I think many problems evolve through human failures.
Computers are more precise, don't have any mood and performance swings, no attention span and no declining attention, they can calculate many variables at one time and with a higher calculation speed...
However, computers can get technical malefunctions, could be disabled by external factors, and they are only as good as their program. And I personally don't know how mature and evolved the AI of these programs already is.
Therefore I think a middle course is the best:
Combine both aspects with each other.
Let a computer reduce the risk of human failure, but let the human pilot be able to overwrite the computer's control.
And if tyrants take me, And throw me in prison, My thoughts will burst free, Like blossoms in season.
Foundations will crumble, The structure will tumble, And free men will cry:
Thoughts are free!