3 ms·

A self-driving car system might be statistically safer than a human drivers in overall conditions, and still be prone to failure in an exceptional situation a p

A self-driving car system might be statistically safer than a human drivers in overall conditions, and still be prone to failure in an exceptional situation a particular driver wouldn't have been. If you have the potential to be that driver in that situation, I could understand feeling the scare quotes are warranted. Which percentile of the population does an automated system need to outperform to be safer rather than "safer"?

JshWright 13y ago
I am (in my own estimation) a very good driver. I drive a manual transmission car, and enjoy using the flexibility that provides in my daily driving. I drive emergency vehicles of various sizes, in a lot of less than ideal conditions (I live in upstate New York, so weather is a big factor in driving conditions). I have no problem believing that a computer will be able to do an objectively better job of driving than I can. It will be able to react more predictably in routine driving, and respond _much_ faster to exceptional situations. In a collision, a modern car is _done_ processing the event (detecting the impact, deciding which safety restraints to deploy, and deploying them) before your body even recognizes that it happened. A computer controlled car will make fewer mistakes and react better in emergencies than a human being will. It has nothing to do with the skill of the driver. It has everything to do with the amount of data that the car can gather (do you have 360 degree radar vision or direct access to sensors built into the car), and the speed with which it can process that data. This technology still has a lot of maturing to do, but it has made amazing strides in the past decade, and I'm confident that it will continue to do that. I, for one, welcome our new robot chauffeurs.
- delinka 13y ago
  Further, once all the cars on the road are automated, they can alert one another to their own changes in routing, mechanical failures, etc, allowing everyone else the chance to reroute and avoid as needed. We could eliminate traffic lights and stop signs and we'll not only get to our destinations safer, but also sooner.
  - JshWright 13y ago
    Not only that, but emergency vehicle will be able preempt roads directly, rather than relying on the 18 year old kid with headphones on to a) recognize the lights and sirens, b) care, and c) do something not stupid (like pull into the way of the fire truck...)
  - hrjet 13y ago
    > they can alert one another Until someone misuses the protocol to gain an advantage or to cause a disruption (assuming it is peer to peer, decentralized). If it is centralized, then um, there are other worries (privacy, control, etc).
- WiseWeasel 13y ago
  So you feel that 100% is easily within reach? As a coder, I can't help but think it'll be a while before we iron out all the bugs in the driving AI, and as a result, we'll see self-driving cars on the roads long before they are able to outperform 100% of human drivers in all situations, or even your average sober, attentive driver in all situations. Even if their reaction processing times are orders of magnitude faster and the sensors give you significantly better awareness, if there are conditions where the sequence of AI responses and priorities are conflicting or wrong, then you'll just be faster and better at getting the wrong solution, sometimes tragically so. I don't doubt that we'll ultimately be much better off with self-driving cars being the rule, but it may be a bumpy road to get there.
  - JshWright 13y ago
    I never said easily. In fact I said the technology still had a lot of maturing to do. I also never said it would be 100% (though I'd dare you to find me any human who would meet that 100% criteria either). I absolutely agree there is still a ways to go before this is mainstream technology. I also agree that we will get there eventually, and the world will be a better place for it.
maxerickson 13y ago
The more interesting question might be the percentile where the auto-driver makes driving safer for everybody (by weeding out the worst drivers). I would guess it is more like 15 or 20 than it is like 50 (but of course I'm guessing).
Planar 13y ago
We (automotive active safety engineers) don't look at it as a percentage-of-inept-drivers, but more like the Trolley Problem[1], that is, overall harm vs benefit. To put it another way: seatbelts, airbags, and heart surgery hurt/kill some people, but the benefit outweights that harm. [1] www.wikipedia.org/wiki/Trolley_problem

JshWright 13y ago

I am (in my own estimation) a very good driver. I drive a manual transmission car, and enjoy using the flexibility that provides in my daily driving. I drive emergency vehicles of various sizes, in a lot of less than ideal conditions (I live in upstate New York, so weather is a big factor in driving conditions). I have no problem believing that a computer will be able to do an objectively better job of driving than I can. It will be able to react more predictably in routine driving, and respond _much_ faster to exceptional situations. In a collision, a modern car is _done_ processing the event (detecting the impact, deciding which safety restraints to deploy, and deploying them) before your body even recognizes that it happened. A computer controlled car will make fewer mistakes and react better in emergencies than a human being will. It has nothing to do with the skill of the driver. It has everything to do with the amount of data that the car can gather (do you have 360 degree radar vision or direct access to sensors built into the car), and the speed with which it can process that data. This technology still has a lot of maturing to do, but it has made amazing strides in the past decade, and I'm confident that it will continue to do that. I, for one, welcome our new robot chauffeurs.

delinka 13y ago
Further, once all the cars on the road are automated, they can alert one another to their own changes in routing, mechanical failures, etc, allowing everyone else the chance to reroute and avoid as needed. We could eliminate traffic lights and stop signs and we'll not only get to our destinations safer, but also sooner.
- JshWright 13y ago
  Not only that, but emergency vehicle will be able preempt roads directly, rather than relying on the 18 year old kid with headphones on to a) recognize the lights and sirens, b) care, and c) do something not stupid (like pull into the way of the fire truck...)
- hrjet 13y ago
  > they can alert one another Until someone misuses the protocol to gain an advantage or to cause a disruption (assuming it is peer to peer, decentralized). If it is centralized, then um, there are other worries (privacy, control, etc).
WiseWeasel 13y ago
So you feel that 100% is easily within reach? As a coder, I can't help but think it'll be a while before we iron out all the bugs in the driving AI, and as a result, we'll see self-driving cars on the roads long before they are able to outperform 100% of human drivers in all situations, or even your average sober, attentive driver in all situations. Even if their reaction processing times are orders of magnitude faster and the sensors give you significantly better awareness, if there are conditions where the sequence of AI responses and priorities are conflicting or wrong, then you'll just be faster and better at getting the wrong solution, sometimes tragically so. I don't doubt that we'll ultimately be much better off with self-driving cars being the rule, but it may be a bumpy road to get there.
- JshWright 13y ago
  I never said easily. In fact I said the technology still had a lot of maturing to do. I also never said it would be 100% (though I'd dare you to find me any human who would meet that 100% criteria either). I absolutely agree there is still a ways to go before this is mainstream technology. I also agree that we will get there eventually, and the world will be a better place for it.

maxerickson 13y ago

The more interesting question might be the percentile where the auto-driver makes driving safer for everybody (by weeding out the worst drivers). I would guess it is more like 15 or 20 than it is like 50 (but of course I'm guessing).

Planar 13y ago

We (automotive active safety engineers) don't look at it as a percentage-of-inept-drivers, but more like the Trolley Problem[1], that is, overall harm vs benefit. To put it another way: seatbelts, airbags, and heart surgery hurt/kill some people, but the benefit outweights that harm. [1] www.wikipedia.org/wiki/Trolley_problem