Because I don’t view it that way as an actual user of the system. I know I am still in full control of the vehicle as is any driver using it. Merely tapping the brakes is hardwired to return control to the driver. The system also has many controls to montior driver attentivness and alert or disable itself if the driver is not attentive. Recently folks have been kicked out of the beta program for lots of reasons so they are monitoring this closely.
I see the argument, but I don't think driver control/supervision is actually the crux of the issue. If Firestone had a completely new tire technology that was mostly ok but had an elevated risk of spontaneously blowing out, I would not be happy if they were doing their beta testing in public roads regardless of who was driving.
The technology being tested seems half baked and not yet ready for testing on a public roads. At the very least, their approach to software updates does not seem like its being sufficiently tested before deployment. This video[1] linked in another thread is an example of what I mean.
Its not clear to me that a general purpose autonomous car can be developed in a vaccum as you mention. Tesla has all the resources in the world to test with their own drivers on closed loops and using simulations however they choose to go this route and open themselves up to significant liability. Given the pace of development of other autonomous cars I think this may be the only viable approach.
Just because you can make steering, braking, and acceleration changes to the car does not mean you are “in control”. The software makes such aggressive decisions that it can become impossible to react, causing massive additional danger to others, such as was the case in the recent crash.