My guess is that the levels of these systems will differ. Some systems will be better than humans (on average), and some systems will be worse. The gamble I mention is that Uber and Tesla are putting too many machines on the road at once without having good priors for their current level of competence.
> The gamble I mention is that Uber and Tesla are putting too many machines on the road at once without having good priors for their current level of competence.
And from whom do you expect this data will come from? How many is too many? Why?
As far as I'm concerned as long as a human can take control at a seconds notice, the risk here is very much overblown.
I haven't seen an instance where a company wasn't taking reasonable precautions and was acting foolishly. I'd love to be proven otherwise...
There's a reason why companies waited until 2016 to start putting these on public roads. And even then they are usually limited like Teslas. They are taking it slowly like responsible engineers...
Now that full data capture and machine learning is being added pretty much every ride is contributing to the quality of these systems.
The problematic part for me was Tesla bragging about safety of autopilot before they had enough significant data. Basically saying "We still don't have any fatalities so he'll yeah we must be safer than human!".
The sad part for skeptics like me is that we couldn't really oppose this kindergarten argumentation. We had to wait for the first fatality (wich happened as you must all know).
So now if I base myself on Tesla own logic, I'm only one death away of saying that Tesla autopilot is far worse than average human driver...
And this actually is gambling with human life.
PS: Noteworthy if theses gamblers are wrong this could easily backfire on the whole autonomous driving industry.
> And from whom do you expect this data will come from?
If they have that much money, they can build their own test city and populate it with volunteers. I'll be thrilled to see Travis Kalanick walk over the mock street 100x per day while the cars are being tested.
> As far as I'm concerned as long as a human can take control at a seconds notice, the risk here is very much overblown.
That worked out splendidly for Air France flight 447 and many others.
We gamble with pedestrian's lives every time we get in a car.
No one has seen how these vehicles perform in real world scenarios. It seems very possible they could perform better than humans on average.