Airplane pilots have plenty of time for fixing any problem that may appear mid-flight. It's because of this that airplanes don't have autopilots landing or taking them off.
And by the way, there's an ongoing discussion on airplane groups about all the new failure modes created by the autopilot-pilot interface. It creates plenty of problems.
There's no inherently intractable reason aircraft "need" human pilots when autopilots are faster, better and make fewer errors. At this stage, an aircraft guider would be able to instruct the aircraft as to what's needed. At some point in the future, AI/ML/DL will replace that role entirely such that the lineman communicates with the aircraft directly. Then, as a further iteration (40-100 years), the lineman is also replaced by robotics on a drone-like platform.
No to the first part. At the risk of sounding flippant, this is one of those cases where the aviation industry really does better than you’d think. Humans do make errors, but, for you to apply ML to fix it, you need to make a lot more errors of all possible types to build robust training data. Sure, it may happen one day, but until then we are dependent on pilots and pilot aids. People are just much faster and more capable at coping with novel, or rare situations.
Regarding drones, I suspect that you are at least partly right. There is a lot of interest already in drones as inspection platforms. Perhaps they will go farther as drones get more commercial aviation clearance.
