> didn't have any real empirical evidence to back any of our claims up. NHTSA publishes the FARS every year. It is their database of every known roadway fatal

> didn't have any real empirical evidence to back any of our claims up. NHTSA publishes the FARS every year. It is their database of every known roadway fatality in the United States. It's free to the public. It's well worth study. Here's an example: Texas has more fatalities than California. Not per-capita, but in total. California has 10m more people living there than Texas. This is not a uniform problem. > Presently human drivers cause an awful number of deaths with cars. Yes, a majority of accidents only involve a single vehicle. It is often the case that a driver exceeds their ability or roadway conditions and kills only themselves. > would drunk drivers still kill more people every year even if we replaced all of the cars with robot ones? Replaced them with cars that have no manual control option. Given everything we've learned about automation, this is a bad idea given our current infrastructure. So, it's highly unlikely that they will have much of an impact for this group. > it seems like we'd be faster at making them safer than preventing humans from making poor decisions. This ignores the nearly 100 years of saftey history that we already have. We have already reduced roadway fatalities by a staggering amount and have created a regulatory framework that has successfully created higher saftey margins for all drivers. We already drive Trillions of miles every year in the US and we only have 36k fatalities. Further: 6k of those are pedestrians and 8k of those are motorcycles. 7k are young men under 24 and 7k were drivers under the influence. There are not a lot of "easy gains" left out there where it's obvious that self-driving cars will have a major impact. > I think in the case of piloting a slow-moving bullet around crowds of humans Which is one thing. However, we're talking about piloting thousands of slow-moving bullets around crowds of humans _and_ each-other. There will not be a single manufacturer of technology, nor will there be a single auto-pilot system. Plus we haven't experienced any automation failures yet that have caused us to establish best-practices around this system. For example, should the autonomous system have full 100% authority over the cars controls? It may be useful in an accident that the car can use full authority to avoid an accident. It may be disastrous in a situation where the car has bad data from it's sensors. So, should we limit control authority to reduce the opportunity for failure here and also leave some margins for a human to override the machine? Will drivers actually understand the complicated and subtle difference? What happens when we need to roll out regulatory changes to autopilot functions? Do we do that all at once? How do we force drivers to do these updates? How do we keep drivers safe now that we've changed a major operational mode of their vehicle? We expect airline pilots to engage in simulations and continual training to ensure that they are on top of these changes, what are we going to do in vehicles? > It did wonders for the aviation industry It wasn't without it's own set of challenges[0]. Upshot of the video is, switching to a lower level of automation is now recommended practice in many situations because it's actually safer. [0]: https://www.youtube.com/watch?v=pN41LvuSz10 https://www.youtube.com/watch?v=pN41LvuSz10