[creshal]

It's also why airplane pilots do much more than they theoretically would have to – under normal conditions, a modern autopilot should be able to handle the whole flight without human supervision.

But then the human won't be reacting fast enough when he is needed to handle an unforeseen problem, so they're kept in the loop and active, just in case.

[prower]

I wonder how many flaps you could possibly check on your Tesla during your commute (i'm being sarcastic).

Also, I could imagine that an airplane is, from a totally theoretical and simplicistic point of view, much easier to manage for a computer, even for simply having a lot of space to maneuver, and knowing constantly what other airplanes are doing.

But driving a car? How can a computer possibily handle a scenario where edge-cases make for a good 85% of each travel, unless you drive in the desert?

[the8472]

It doesn't have to be a 1:1 replacement. The computer has some advantages that can make up its lack of brains by having lower latency, not getting distracted, better sensors, better maths when it comes to motion ("objects in the mirror may be closer than they appear") and more training than a human will ever experience.

[prower]

I totally agree. But also, critically, an auto-pilot is precisely expected to have "a brain", it's implicit in the name.

Maybe we should just agree that auto-pilot, in its common interpretation, is currently just a sci-fi concept, and communicate more honestly what a computer inside a car can't and can do.

Apparently, dodging weird obstacles is something it struggles with. Which is a problem, since all driving is, essentially, dodging weird obstacles.