Whenever I get into an argument with someone about the merits of self-driving cars, I point out that we're pretty much there already with flying.
Two other points.
1. Driving happens "in 1-and-a-bit dimensions" (i.e., cars are on roads, which are 1-dimensional plus a bit of width) and aviation happens "in 2-and-a-half dimensions" (i.e., planes are in the air, fairly close to earth, but with quite a lot of flexibility as to height), which means that there's far, far more space for planes than for cars.
If all the world's drivers were completely unable to tell where other vehicles were, all the roads would be filled with crashed cars within seconds. If all the world's airline pilots were completely unable to tell where other planes were, the effect would be minimal. Except that there'd be a bit of danger at takeoff and landing -- which, not altogether by coincidence, are two other things that are best done by human pilots rather than unaided machinery.
2. Cars operate in a world where there are also pedestrians, buildings to crash into, etc. Planes, not so much.
It seems to me that, to a very good first approximation, none of the problems self-driving cars need to solve apply to planes in sufficiently similar ways for "we can do it with planes, so why not with cars?" to be a good argument.
There's nothing 'simple' about landing an airplane in water that is not designed for it, especially a glider.
Most computer guys don't realize this, but you can't save the state of a flight. Most decisions are mostly permanent once made. And some, like for instance all decisions made on an aircraft with no engines, are completely irreversible. So if you don't make the right choice on the first try, you have to live with the wrong choice. At low altitudes under visual flight, a pilot makes multiple such decisions a minute. In an emergency, this number would be a lot higher; on the order of once every few seconds.