4 ms·

Why not just have a human in the car??? Is equipping a car with a 5g modem so that you can pay 2$/hr an hour to someone in SEA to drive your car really the hori

by airza 4y ago

Why not just have a human in the car??? Is equipping a car with a 5g modem so that you can pay 2$/hr an hour to someone in SEA to drive your car really the horizon of human achievement?

luckylion 4y ago
You'd only need a human while the car is driving. If the human needs to be physically in the car, they couldn't just switch to a different car when this one has been parked, and you'd need to organize their own travel when they take a car somewhere remote, or need to fetch it from there. This would still be massively cheaper even if you pay the driver whatever you'd pay them locally for driving in person, simply because of less downtime, e.g. while a truck is being loaded.
- airza 4y ago
  So it’s being a driver without any of the downtime of a driver? That’s somehow even worse!
pcthrowaway 4y ago
Let's say you have 1000 self-driving long-haul trips that have an accident rate of anywhere from 0.1*A to 10*A (where A is the rate of accidents with human drivers) with just self-driving. Let's say 100% of the time there is an accident, the self-driving system is able to determine with at least 1 second of notice that human intervention could make the upcoming segment of the trip safer, and such flagged sections constitute less than 0.1% of the time the vehicle would otherwise be in operation. Let's also say a highly trained team of 10 people, monitoring that fleet of 1000 vehicles, selected for fast reaction, and with adequate training could theoretically prevent 99% of at-fault incidents, bringing it down to a range of 0.001*A to 0.1*A. I don't know if this would actually play out, but if the above assumptions were found to be accurate, then it'd be hard to make an argument for having human drivers in the vehicle when this system at scale could reduce at-fault accidents to <10% of the current rate Think Mechanical Turk meets Lambda for improving self-driving car safety

by airza 4y ago

Why not just have a human in the car??? Is equipping a car with a 5g modem so that you can pay 2$/hr an hour to someone in SEA to drive your car really the horizon of human achievement?

luckylion 4y ago

You'd only need a human while the car is driving. If the human needs to be physically in the car, they couldn't just switch to a different car when this one has been parked, and you'd need to organize their own travel when they take a car somewhere remote, or need to fetch it from there. This would still be massively cheaper even if you pay the driver whatever you'd pay them locally for driving in person, simply because of less downtime, e.g. while a truck is being loaded.

airza 4y ago
So it’s being a driver without any of the downtime of a driver? That’s somehow even worse!

pcthrowaway 4y ago

Let's say you have 1000 self-driving long-haul trips that have an accident rate of anywhere from 0.1*A to 10*A (where A is the rate of accidents with human drivers) with just self-driving. Let's say 100% of the time there is an accident, the self-driving system is able to determine with at least 1 second of notice that human intervention could make the upcoming segment of the trip safer, and such flagged sections constitute less than 0.1% of the time the vehicle would otherwise be in operation. Let's also say a highly trained team of 10 people, monitoring that fleet of 1000 vehicles, selected for fast reaction, and with adequate training could theoretically prevent 99% of at-fault incidents, bringing it down to a range of 0.001*A to 0.1*A. I don't know if this would actually play out, but if the above assumptions were found to be accurate, then it'd be hard to make an argument for having human drivers in the vehicle when this system at scale could reduce at-fault accidents to <10% of the current rate Think Mechanical Turk meets Lambda for improving self-driving car safety