There are videos of the Autopilot feature attempting to swerve into oncoming traffic. It does have the potential to kill people, it likely will in wide usage just by raw statistics, and it's likely to eventually fall under regulation of some sort. That doesn't mean it's necessarily a bad feature, but pushing self-driving car functionality as a "beta" to the public is, by at least some definitions and to some extent, "Elon Musk playing with people's lives".
This is not born out by the currently available facts.
Tesla quoted deaths-per-autopilot-mile versus overall deaths-per-mile for all cars on the road. This is very misleading information, for a number of reasons.
1. Teslas are high-end luxury cars and thus have better safety features than the lower end of the market.
2. Teslas are likely to be newer and better maintained than the average car on the road. The average age of a car on the road is 11.4 years (https://www.google.com/search?q=average+age+of+cars+on+the+r...) - there are ZERO Tesla cars that old.
3. Autopilot is disabled on residential streets and those with no center divider (http://www.newsmax.com/TheWire/tesla-autopilot-videos-restri...), which means it's driving in more predictable situations than average. You're also unlikely to have it on when there are things like icy/snowy conditions.
4. Autopilot will hand over control when it can't handle something. If it does this suddenly, the subsequent crash may not count in Musk's "autopilot miles" statistic.
Most drivers will take over and disengage Autopilot in doing so, so your point remains, but I wanted to make it clear that the car isn't just doing a "Jesus^H^H^H^H^HDriver take the wheel!" thing.
Is it when the alarm sounds? When it begins to slow? When it turns on the hazards? When it has come to a complete stop? How would being rear-rended during this process be counted?
3 this is valid point, but it's significance is diminishing as autopilot is accumulating death-free miles. also does tesla not deal with icy/snowy conditions in Norway?
4 is pure speculation.
#1 and #2 are not at all irrelevant. This sort of technology will make its way into lower-end cars in time. Having actual data on how safe it is, with other variables accounted for, is going to be extremely critical.
It's like claiming racecar driving is really safe because not many people die doing it. That's not actually because it's safe - it's because the drivers are very well trained, with very good safety equipment so they survive crashes.
Teslas are safer than average cars. The autopilot may not be.
> 3 this is valid point, but it's significance is diminishing as autopilot is accumulating death-free miles.
It is certain statistically to have more fatal accidents as it accrues more miles. The underlying issue - that "autopilot miles" are safer miles than "all miles" - will remain. Its accident tallies need to be compared statistically with miles with human drivers in similar road conditions. This will likely require a research study.
> also does tesla not deal with icy/snowy conditions in Norway?
I couldn't say for certain, but it apparently will shut down in moderate rain (http://www.teslarati.com/tesla-autopilot-limitations-heavy-r...) so I'd guess snow would have similar impact on its visibility and functionality.
> 4 is pure speculation.
#4 is an important question.
http://www.iihs.org/iihs/ratings/TSP-List
Per http://www.torquenews.com/1083/why-tesla-s-2014-model-s-cann... - has some interesting comments, too.