Statistically computers can perform better given good programming or given extremely large data samples which they can build patterns out of. However, if you are still stuck with the mind-set that computers are more "capable" than the human mind you really ought to learn what the word "compute" means. It doesn't imply an innate ability to do something a human can't. To compute reality (as one would expect a truly automated passenger jet aircraft to do) accurately and successfully you need an interface to the real world. The issue is not with computing power, rather the interface that enables the computing power to be used.
If that interface is unreliable; uneconomical to build and maintain; liable to interference; can be hijacked; breaks when it rains or whatever then you have a fundamental problem in delivering reliable end to end automation. That's what I was talking about when I was saying a billion years of evolution cannot be outdone by something we are going to build, at least in this century.
The arguments involving war machines and UAVs are wrong. They don't carry human payload for starters and are much lighter by comparison. If one of them crashes no big deal, we build more. If people get hurt because they crashed, no big deal they were probably enemies!
Here's an interesting statistic. Today's level of AI has only just managed to mimic the intelligence of a fruit fly and AI is a field being advanced for at least 50 years now. From that is posed a question to the yes brigade, what would you consider to be a safe automated passenger jet? Is it one that can think as well as human (i.e. have the same level of intelligence) or one that can think a billion times faster given the same (known problems)?
Amazing that this just made front page on the BBC:
BBC News - Stephen Hawking warns artificial intelligence could end mankind