Thing is, you can test a program and fix it, the human not.
DenverPilot addressed the second point. I'll address the first.
Only the very simplest bits of software code can be tested exhaustively.
Every "branch" (i.e., decision) in the code doubles the number of test cases. Now, with a million lines of code (say, FMS -- what you're proposing would be substantially larger), you might expect a few hundred decisions. Quick, what's 2**100? At a rate of one test per nanosecond (ridiculously fast....), how long would it take to test this code? How does this compare to the age of the universe (13.7 billion years -- about 5x10**17 sec)?
You can "white box" test to reduce this, but then it's no longer exhaustive testing, as design assumptions must be reused to make the tests.
And testing is very, very expensive, so anyone trying to make a profit will attempt to minimize them.
It gets worse with event driven code, which almost all control systems have a flavor of. That's because you get to consider time ordering. For two events, they can occur in two orders, or simultaneously. Leaving simultaneity out means you program in an uncontrolled failure if it ever happens (and it's much more likely than you might think, as all computer systems discretize time).
The technology to subvert a flight through the use of computers and remote coms is much more difficult seems to me.
