The software can only act on information fed by other systems, sensors, signals. On a complex installation there are thousands of inputs/outputs, some virtual (internal software values) and some generated physically. For some of these I/O, under certain failure modes, the computer will not be able to tell if a signal or variable is true, or in some cases which of multiple signals is giving the 'real life' value if there is a mismatch. In the case of a major fault developing (e.g. Control panel fire) the computer may lose some or all of its I/O.
B&2, that makes a computer no different than a human - we all function based on I/O - and if the input is corrupted, we often do something stupid (see Birgenair), and if the flight deck is on fire, I wouldn't put good odds on any human putting that airplane on the ground safely. The advantage a computer has is it can process infinitely more inputs than any human. If airspeed is corrupted or questionable, the computer can, in a fraction of a second, compare GPS, Doppler, different air data, pitch, etc. and determine which inputs are valid. High traffic environment? How many independent aircraft can the typical pilot track in their immediate airspace (without missing any)? Five, ten, if they're really, really good, twenty? With a computer you're talking thousands, and it can use multiple independent inputs to make sure it doesn't miss one. All engine power loss? A computer can accurately calculate the best glide airspeed and max range to a potential landing site, and if no landing site is in range evaluate options as to where to put the aircraft down that will minimize casualties, and do it in a fraction of a second. The programing to do that is not that hard.
I work engines - if the software in your FADEC is wrong, it can cause every engine on the aircraft to do the same wrong thing at exactly the same time (e.g. shutdown at 500 ft. AGL after takeoff). Yet I'm not hearing people say they won't fly on an aircraft with FADEC engines. Now, because the aircraft in question can easily fly with one engine shutdown, we don't always try to keep the engine running for certain fault conditions so we don't end up with something worse than a shutdown (e.g. uncontrollable high thrust). We could easily design the engine control to try to keep the engine running no matter what, but we've made a conscious decision to allow the engine to shutdown or default to idle as that's considered to be the safer option.
Today, the design philosophy when things fail is to turn it over to the pilot because he/she knows more about the exact situation than the designers do. I've had some pretty heated discussions on how to deal with various failures - maintaining that we don't to automatically take certain actions because it can mislead the pilots as to what is really wrong (even if taking that action makes sense). But that doesn't always turn out well either (AF 447).
Humans still have one big advantage - being able to react to new and/or unknown situations (although they don't always get them right either). But true, unimaginable 'new' situations are really not that common.
I'm not suggesting it'll happen in the next 10 or 20 years, but it will happen. If you'd told a WWII pilot in late 1945 that, within his lifetime, he'd be able to fly on a computer controlled airplane that could travel 8000 miles at Mach 0.85 without refueling he'd think you were crazy. Computer capability is still growing at exponential rates - basically doubling every 18-24 months (that's showing signs of slowing down, but not by much). Human capabilities, not so much.
