The number of miles driven autonomously doesn't really matter if drivers have to intervene even for a second to resolve an issue. If the car is driving by itself 99% of the time, that's not good enough for the "share an autonomous car" use case. It's no good at 99.9% either. Or 99.99%. It's a lot more interesting if google's cars actually had no steering wheel or way of intervening.
Because that's what we're talking about when we talk about the price reduction of a fleet of autonomous cars: the car driving to your house, completely on its own, with nobody in the car. 99.9% isn't good enough, and we're not even close to 99.9%. The difficulty of the last remaining difficult problems will only increase as we approach 100%, too.
I think completely driverless cars are a problem on a similar scale as AI or nuclear fusion, but with the timescales decreased... we're probably going to be always perpetually "five years out" from having them (as opposed to fusion's perpetual "30 years"), even though we're making improvements all the time. We just don't understand the scale of the problems we haven't solved yet.
The thing with autonomous ride sharing though, is that the car has to make it from its last stop to your home on its own. And there's any number of scenarios that can happen on its way... a traffic light could lose power and turn into a 4-way stop, for instance. Or a big event at a large venue where hundreds or thousands of pedestrians are crossing the street and you kinda have to inch forward and wait for some of them to wave you on to go forward (basically everybody jaywalks at that point.)
There's a million different 0.01% scenarios that could happen, and if you ramp the usage of a really really good autonomous car up to scale, you'll see them every day. Unless it can handle all of them without intervention, you simply can't ditch the driver, or you'd face huge PR nightmares when you either hit a pedestrian or you block an intersection and cause huge traffic in a major metropolitan area ("Mayor bans autonomous cars due to one too many gridlocks caused by them", etc.)
I really think it's down to the PR issues and the fact that one too many mistakes can really kill the idea, that companies will be cautious to just not need a driver altogether. And for autonomous ride sharing, you must be able to do nearly half your miles without a driver. Hence why that last .01% is still crucial.
That is the thing with driving - it's amongst other drivers, pedestrians, especially children. There are so few use cases where 99.99% is enough. I can only think of one, highway from A to B - how you get to and from the A and B respectively is not by selfdriving cars.
Because that's what we're talking about when we talk about the price reduction of a fleet of autonomous cars: the car driving to your house, completely on its own, with nobody in the car. 99.9% isn't good enough, and we're not even close to 99.9%. The difficulty of the last remaining difficult problems will only increase as we approach 100%, too.
I think completely driverless cars are a problem on a similar scale as AI or nuclear fusion, but with the timescales decreased... we're probably going to be always perpetually "five years out" from having them (as opposed to fusion's perpetual "30 years"), even though we're making improvements all the time. We just don't understand the scale of the problems we haven't solved yet.