I wouldn't have a problem with that if it's in a framework where they would put their money where their mouth is, and accept liability for everything the car does incorrectly while using vision.
As we are not there yet, we can either opt not to use it, or accept the experiment as guinea pigs, retaining full accountability for a system we do not know the limitations of, having to make split second corrective decisions depending on the unpredictable behaviour of the car, that may or may not change with every incremental software update... It's not ideal.