I agree with much of what you've written. I think the main difference between our views is in who has the onus of proof to make a change in policy. You seem to be skeptical and want Tesla to prove that Autopilot is safer; whereas I see Autopilot's design intuitively as a safety improvement, and would want data to the contrary to withhold it from the market.

While we don't have a statistically significant data set to pine over, I do think we have evidence that "a driver who [falls asleep at the wheel or texts and drives] will fare better with Autopilot than without it." I use Autopilot on my way to and from work each day. It's pretty typical for AP to handle a ten minute stretch with no input from me. For any given minute of driving, there is a very high probability that I haven't given any input to the car. It follows then, that for any given minute on Autopilot during that stretch, I'd probably be fine if I were not paying attention. This stands in contrast to the alternative: if I gave no input to the car for an arbitrary minute without Autopilot, there is virtually no chance that I wouldn't cause an accident. This is where my intuition derives from.

> The main difference between AP and other safety measures is that, based on Tesla's defenses so far, there is literally no situation in which AP should be assigned blame, because it's ultimately the driver's responsibility. Whereas when an airbag fails to deploy, or just randomly blows up in a driver's face, there can be agreement that the airbag "failed".

I understand why this sounds like passing the buck, but I think this is the right policy (for the driving public's sake). People need to be responsible for how they use any L1 or L2 system, period. Again, I would liken it to Automatic Emergency Braking. There is no situation where a failure of AEB to brake would render a driver not responsible for an accident. The system adds a layer of protection, but it does not make any guarantees. I suppose if AEB or Autopilot acted in such a way that it overrode driver input -- say AEB braked hard on the highway and ignored the gas pedal input -- then you could blame AEB. And I'd say the same for Autopilot -- if it e.g. veered toward an object and overpowered a human trying to avoid it, I would blame AP for that. But as far as failures of logic, I think the driver owns that entirely until L3 systems are available.

> the question for overall safety is: does AP result in a net increase of accidents because drivers are too over-confident?

This is a good question, and I hope Tesla releases its data as promised so we can put it to rest. My own experience is that Autopilot makes it much easier for me to pay attention to my surroundings because I don't get fatigued and I'm less stressed than when I'm driving manually. I recognize that my experiences could be atypical, but having seen hundreds of similar sentiments on TMC over the years (and very few counters), it would take data to convince me otherwise.