But to reach that point, we've got to accept a learning/development process that will involve some risks and some annoyance.
Uhh, about that...
> Tesla allows self-driving cars to break speed limit, again
- https://www.theguardian.com/technology/2017/jan/16/tesla-all...
It's the same with Tesla wanting to allow rolling stops at stop signs – not the letter of the law, but it's what humans expect. In many instances you are more likely to cause a minor accident (being rear-ended) by coming to a complete stop at a stop sign than the risk of causing and accident by rolling through an intersection that clearly has nobody else in it.
Won't be a problem once most cars on the road are driverless.
The time to find out your car does not stop for emergency vehicles, is not during an emergency. Yet here we are
How about all those new drivers hitting the road for the very first time? Are you suggesting we should banish them from public roads until they've got a few hundred hours under their belt?
How are they going to get freeway practice?
That'll happen only when we have solved the general intelligence problem. Driving on roads is a cooperative process, it requires a theory of mind, and many people are sick of making allowances for experimental technology that is inherently incapable of performing as intended. It's the equivalence of goldmaking alchemists.
Meanwhile there's us motorcyclists - I trust robotaxis to even estimate distances as much as I trust drunk drivers or teenagers on their second driving lessons.
Banish that shit to test tracks.
Don't get me wrong, this is an impressive feat—that sometimes crashes into firetrucks.
Until they stop crashing into firetrucks, which are rather large, bright red, and pretty hard to crash into, the idea that they are going to be better than most drivers is speculation.
However it is very likely that self driving is in some sense AGI complete - in other words to get to the goal may require something just as generally intelligent as humans, not least as driving in a city is a social interaction problem as much as a routing problem.
Have most people just not encountered confusing road conditions?
A couple weeks ago, I was in West Virginia bobbing up and down hills around hairpin turns on loose gravel. The whole time, I was thinking to myself "The first autonomous vehicle that attempts this route is going straight into the ravine."
One time evacuating from a hurricane, the route I found required me to drive through an open field. How well does LIDAR cope with tall grass?
Another time, I'm headed north on I-95 and there's a several car pileup. Police divert all traffic off the nearest exit and close the highway with a few road flares. I'd wager anything an AV would blow right past the flares and wreak havoc on the scene of the accident.
Until AI can cope with completely novel scenarios it was never trained for, it's going to be prone to catastrophic failure.
Personally, I don't expect to see it in our lifetimes.
They slow down. Tesla's dumbest form of automation will slow down more than a human when encountering a turn.
> How well does LIDAR cope with tall grass?
It stops and demands help. While car automation leans on GPS/nav, it won't blindly follow it -- similar to (almost all) humans.
I wouldn't want to risk anything near a vehicle accident on the road, though. But the situation is not _as_ bad as you might think.