A machine that has actually failed in some way has proven that it is not "100x more reliable" than anything. It is broken. It is no longer a matter of probabilities.
No it hasn't. A machine that will perform a given task correctly 9999/10000 times is 100x more reliable than a human that will only perform that same task correctly 9900/10000 times. A rare combination of factors that make the machine fail don't make it absolutely broken, any more than the human is "broken and no longer a matter of probabilities" if they make a mistake sometimes.
Believing that machines can be "aware" is the failure here.
I'm not sure what metaphysical definition you're using for 'aware'. When a control system (biological or not) gathers information via a sensor and processes that information in a way that can affect the system's behaviour, then it's 'aware' of the thing that sensor measures.
As for the straw-man argument about the stroke victim, consider this alternate scenario: You've got a passenger in the back seat of your conventional, manual-controlled car and you're heading to the hospital. You have a seizure and lose vision in both eyes. You're "aware" that your video sensors have failed, and refuse to drive further. Would I, the passenger in the back seat, prefer to be perfectly safe with the car pulled over, or would I prefer that you kept going despite failed sensors (maybe with me shouting out "left a bit!" "right a bit!" to try and guide you)?
I wonder if population studies have been done, how does the ecosystem recover after the algae bloom?
I haven't checked either, but I'd guess that the water will gradually absorb oxygen from the air until it reaches a livable level, at which point the surrounding ocean ecosystem will recolonize it.
The victim is already victimized and will not be un-victimized.
The victim continues to be victimised for as long as the images are public, because they will continue to be confronted with the images (causing them to relive the trauma) and because anyone they meet has potentially seen the images and will treat them differently because of it.
what are the odds that you'll remain alert and aware of the surrounding traffic after the 100th commute where it proved completely unnecessary to do so?
Spot on. It doesn't improve safety in any way shape or form. It's just a liability dodge. So far, vehicle manufacturers have been able to offload responsibility for crashes onto the drivers involved unless it's provable that the car was engineered wrongly.
Fully autonomous vehicles are scary for manufacturers because they potentially shift all liability to the manufacturer. This is made worse by the fact that, while people are willing to accept "human error" from a human driver, they become outraged if a machine makes a mistake, even if the machine is 100x more reliable than a human. This is a mindset that will have to change as machines become more aware of their surroundings and start making higher level decisions.
There are two ways to write error-free programs; only the third one works.