http://en.wikipedia.org/wiki/Deadstick_landing
When you're in a truck everything seems to be OK until some kind of non-obvious corner case comes up and BOOM! "you" (the truck) just crashed into a car and killed a family.
Not a job I'd want since you probably have all the liability for the problems and none of the time to react that you'd want. Watching a truck drive itself -- if you're on call to jump in and try and avert a disaster -- is probably more stressful than just driving it yourself.
Patrick Smith (ask the pilot) has a number of good criticisms [0][1] of this "1%" trope.
[0]http://www.nytimes.com/2015/04/10/opinion/why-pilots-still-m...
Assuming the control system is smart enough to cope with changes in control authority, it's got to be better - it has much faster reactions and direct access to the tyre pressure data.
A system where the autopilot disengages and gives up control without having confirmation that the human has taken over (by giving a control input) is an obvious danger though.
That's my exact point - you can't hard code in every single possible problem. And in case autopilot doesn't have sufficient data or programming to handle the situation it has no other choice but relinquish control to the driver, at a moments notice. Of course it could always try breaking with full force, but a) breaks might be damaged b) breaking is actually a bad idea, if any of the wheels are on a soft surface(like in my example, after a tyre blowout the truck has moved closed enough to the edge that wheels on one side are now on grass - applying maximum breaking force in this situation will actually tip over the truck, the correct solution is to keep going until the truck slows down by itself). The truck might not know what surface it's on or where it's going(say a bird hit the laser sensor on top), so the autopilot would fail spectacularly in this situation, which would not be very difficult for an experienced driver to handle.
The truck should have a notably better feel for the surface than a human.
>say a bird hit the laser sensor on top
Say a bird hits the human driver in the face?
In case of trucks, I've always liked the idea of a convoy with one manned truck and several automated drones - seems like a good compromise.
Well no - the problem with problems is that you can never predict all of them, and complete autopilot would either need to exhibit strong AI to deal with all(not possible right now) or have them all coded in(also not possible). Not to mention that complete power loss can, and has happened before - even if the steering still has power, without a computer running it won't do much good without someone who can move the stick. Or look at the landing on the Hudson River - would any computer ever decide to try it? I think the answer is no. If we had a ridiculously good AI, I think it would decide to crash land the plane somewhere with no people. River has people. There were bridges in the way. AI might not be 100% capable of gliding the plane with no engine trust and broken sensors -it would have simply not tried that landing. That plane only landed safely because the captain had experience with gliders and knew what to do.
It's not like all pilots without that much glider experience are 'unsafe' because they can't recover in that extremely particular set of circumstances.
Even if we ignore the possibility of remote control, you don't need an autopilot that's better in every way than every existing pilot. You just need one that's better than the average pilot.
Let's use a different argument, just for the sake of it. Let's assume there exists a medical device, which never fails on its own, but due to operator errors 1/10000 patients who use this machine die.
Now someone makes a fully automated machine,but due to problems with programming it kills 1/100000 patients it treats(completely on random). Even though it's 10x better than your average operator, such machine would be banned from the market quicker than you can say "litigation". In our society human errors are much more easily forgiven than computer errors, because everyone believes that they could have been prevented. And therefore, a truck autopilot wouldn't need to be just better than your average driver - it would need to be absolutely perfect. And I don't think we are ready for perfect yet.
Until something[0] goes catastrophically wrong. The FMS on that flight threw so many errors that it couldn't compute landing performance metrics. In other words, because it couldn't handle the sheer number of aircraft system failures, it couldn't tell the pilots if the plane would be able to land on the runway and not overshoot it. The pilots hand to manually select which errors to override in order to have the system run its calculations.