We're not as far away from that day as some of you seem to think.
Airbus fly-by-wire flight controls can 'overrule' pilot inputs if it determines what the pilot wants to do is 'unsafe'
(Boeing FBW is somewhat less aggressive about it relative to Airbus, but also can overrule certain pilot inputs). To a large extent, on the newest airplanes the pilots are basically there in case something goes wrong with the automatics.
40 years ago, accident causes were more or less evenly split between pilot error and mechanical failure. Today, mechanical failure has faded into the background - pilot error is the predominate cause (and most of those are CFIT). Early on in the EROPS/ETOPS transition, the safety guys did a lot of work analyzing accidents where engine failure was involved. What they found was that it was rarely an engine failure that caused an accident - it was pilots doing something stupid in response to an engine failure.
Today, most automatic systems on aircraft are designed to default to the pilot when things go wrong - the rational being that the guy (or gal) driving knows more about what needs to happen in that particular situation then some programer sitting in an office years earlier. But to an increasing extent we're finding that 'being there' isn't the same thing as 'being aware' (Air France A330 comes to mind). Remember Moore's law (which should really be Moore's observation) - computing power will double ever 18 months - while humans have pretty much stagnated. Fully autonomous cars are in development, and self-driving cars have lapped racetracks faster than professional race drivers could do in the same car
We won't get there for many years, but the day
will come when letting a human decide is demonstrably less safe than letting the automatics decide. When we get to that point, it would be irresponsible to let humans drive with hundreds of people in the back.
As for costs - back around 1980, when Boeing was developing the 757/767 - the FAA came out with their finding that a 3 crew flight deck was no safer than a properly designed 2 crew flight deck. The launch customers immediately came in and told Boeing to change the 757/767 flight deck to a 2 crew design. Boeing said 'OK, but it'll cost you another million dollars per airplane' (the first 7 or 8 767's were actually built with 3 crew flight decks and needed to be retrofit for 2 crew and EICAS). The airlines response was 'no problem, we'll save at least a $million per airplane each year in crew costs going to 2 crew'. That was in 1980 dollars (and flight engineers didn't make as much as pilots)
BTW, it's my understanding that the systems for full autonomous trains have been around for years - that they haven't been adopted for widespread use has more to do with the power of the unions and public perceptions than any technical limitations.