3 ms·

> The details of their deaths are no important. Saving 25 lives per day is clearly the right thing to do. I think it's clear that if we were talking about savi

> The details of their deaths are no important. Saving 25 lives per day is clearly the right thing to do. I think it's clear that if we were talking about saving 125 people 90 years old or older or 100 children, the clear choice is to save the children, from almost any rational point of view. Even just from a societal health or utility standpoint, children are worth more than people that are statistically close to death anyway at a societal level. If that premise is acceptable, and if we accept that switching to self driving might change the demographics of who dies in accidents, then it's important to quantify not just whether it saves more people or not, but what people, in case that shifts matters. > Hint: Imagine all of these people are your family and friends. Do you want to keep having 125 of them die every day, or only 100? I would rather lose both my parents than one of my children. I don't know a single parent that would feel otherwise. Additionally, there's also the possibility that like adding extra lanes of traffic to a freeway can make traffic jams worse, self driving cars could increase deaths per capita while reducing deaths per trip, by promoting a lot more car travel (by either allowing current non drivers to drive or increasing time current drivers spend in the car).

grecy 3y ago
> I think it's clear that if we were talking about saving 125 people 90 years old or older or 100 children, the clear choice is to save the children, from almost any rational point of view. Even just from a societal health or utility standpoint, children are worth more than people that are statistically close to death anyway at a societal level. That assumes that the people currently being killed on the roads are "less valuable" than the people that would be killed on the roads by autonomous systems. There is no evidence to prove that, it's purely hypothetical.
- kbenson 3y ago
  Yes, there is no evidence. The point I was trying to make is that gathering that evidence and knowing what is happening is important. It's not enough to just assume any drop in deaths is automatically better than previously, when what we're talking about is a drastic change in how the overall system operates. Im not arguing that autonomous cars will be worse, I'm just arguing there is the possibility for them to be worse in non-obvious ways, and you prior arguments seemed to ignore that. We should not be so blinded by a single metric that we fail to understand what's really going on. To are many cases of that in technological advances, legislative change, and societal shifts, that we aren't aware of until long after the additional damage has been done.
simondotau 3y ago
That's an absurd concern. Children are already the most at-risk on our roads right now, with humans behind the wheel. We're already living your nightmare. The rise of autonomous vehicles isn't going to increase the likelihood of basketballs accidentally bouncing into the street. Nor will it cause a wild swing in the demographic mix of people using pedestrian crossings. By the time they're widely deployed, autonomous vehicles aren't going to be a little bit safer than the average human driver. They will have to be superior by at least one order of magnitude. As in, the number of deaths per million miles at least 10X lower than the human average. I think one order of magnitude is the least we should demand. I would consider it a rather low bar. I suspect that it'd need to be more like two orders of magnitude in order to retain sufficient community consent. So if that's the threshold of acceptability, then we'd be talking about slashing deaths by 90% or possibly even 99%. Either way, autonomous vehicles could kill children exclusively and they'd still be safer around children than human drivers.
- grecy 3y ago
  > I think one order of magnitude is the least we should demand. I would consider it a rather low bar. I suspect that it'd need to be more like two orders of magnitude in order to retain sufficient community consent. So right now we have 125 deaths every single day on the roads in the US. Same today. Same tomorrow. 45,625 every single year. Year after year. You're saying it would need to be < 12.5 per day to be acceptable, but really we should aim for < ~2 per day. So that means if we invent some system that reduces the number of deaths PER DAY from 125 down to, let's say, 30, you think we SHOULD NOT implement it, because that's not enough of an improvement? Let's say that system is the best we can ever make, but we decide to never turn it on, because it's not a whole order of magnitude better. You're choosing to kill an additional 34,675 people per year, because a reduction in deaths from 125 per day down to 30 per day is "not good enough".
  - simondotau 3y ago
    I’m saying that 10x should be a low expectation for a system that doesn’t get distracted, doesn’t get drunk, doesn’t get drowsy, doesn’t get angry, doesn’t get a bruised ego, doesn’t have a TikTok account, and can look in every direction at every moment. But if the best we can do is a 9x improvement of course we should still deploy it. The question is whether society will accept it. I’m not sure it would. I’m not sure it would accept a system that reduced deaths by two orders of magnitude, or 100x. Imagine for a moment, Tesla fully cracks the FSD nut, and every car on the road is replaced with a self driving Tesla, 100 times safer than humans. “One person is dying at the hands of the pedestrian butcher Elon Musk every single day!” the headlines will cry.
friend_and_foe 3y ago
While I agree with you, my hypothetical scenario above was deliberately constructed to avoid ascribing worth to distinct people, and only measuring how much of the outcome was the result of their negligent or deliberate behavior. We can come to the same conclusions while still assuming all lives have equal worth. All we have to do is determine where the fault lies. If a person jumps out in front of my car because they want to die, that's not my fault. If I run a stop sign and t bone a person it is. The same if I'm a computer in the trunk of a Tesla, and we want to be sure that if we are going to construct an environment where machines call the shots that they're less dangerous, so we should be sure that at the very least they're causing less accidents than people, not simply that they're participating in less accidents. Just the number of deaths doesn't give us this measure, it doesn't actually tell us how safe the machine is. What we ultimately want is primarily that the machine is less often at fault than a human driver, secondarily but very importantly that the machine avoids harming others even when those others are negligent and at fault, and also can avoid harming people in situations deliberately caused by others, particularly if those harmed weren't those causing the situation.

grecy 3y ago

> I think it's clear that if we were talking about saving 125 people 90 years old or older or 100 children, the clear choice is to save the children, from almost any rational point of view. Even just from a societal health or utility standpoint, children are worth more than people that are statistically close to death anyway at a societal level. That assumes that the people currently being killed on the roads are "less valuable" than the people that would be killed on the roads by autonomous systems. There is no evidence to prove that, it's purely hypothetical.

kbenson 3y ago
Yes, there is no evidence. The point I was trying to make is that gathering that evidence and knowing what is happening is important. It's not enough to just assume any drop in deaths is automatically better than previously, when what we're talking about is a drastic change in how the overall system operates. Im not arguing that autonomous cars will be worse, I'm just arguing there is the possibility for them to be worse in non-obvious ways, and you prior arguments seemed to ignore that. We should not be so blinded by a single metric that we fail to understand what's really going on. To are many cases of that in technological advances, legislative change, and societal shifts, that we aren't aware of until long after the additional damage has been done.

simondotau 3y ago

That's an absurd concern. Children are already the most at-risk on our roads right now, with humans behind the wheel. We're already living your nightmare. The rise of autonomous vehicles isn't going to increase the likelihood of basketballs accidentally bouncing into the street. Nor will it cause a wild swing in the demographic mix of people using pedestrian crossings. By the time they're widely deployed, autonomous vehicles aren't going to be a little bit safer than the average human driver. They will have to be superior by at least one order of magnitude. As in, the number of deaths per million miles at least 10X lower than the human average. I think one order of magnitude is the least we should demand. I would consider it a rather low bar. I suspect that it'd need to be more like two orders of magnitude in order to retain sufficient community consent. So if that's the threshold of acceptability, then we'd be talking about slashing deaths by 90% or possibly even 99%. Either way, autonomous vehicles could kill children exclusively and they'd still be safer around children than human drivers.

grecy 3y ago
> I think one order of magnitude is the least we should demand. I would consider it a rather low bar. I suspect that it'd need to be more like two orders of magnitude in order to retain sufficient community consent. So right now we have 125 deaths every single day on the roads in the US. Same today. Same tomorrow. 45,625 every single year. Year after year. You're saying it would need to be < 12.5 per day to be acceptable, but really we should aim for < ~2 per day. So that means if we invent some system that reduces the number of deaths PER DAY from 125 down to, let's say, 30, you think we SHOULD NOT implement it, because that's not enough of an improvement? Let's say that system is the best we can ever make, but we decide to never turn it on, because it's not a whole order of magnitude better. You're choosing to kill an additional 34,675 people per year, because a reduction in deaths from 125 per day down to 30 per day is "not good enough".
- simondotau 3y ago
  I’m saying that 10x should be a low expectation for a system that doesn’t get distracted, doesn’t get drunk, doesn’t get drowsy, doesn’t get angry, doesn’t get a bruised ego, doesn’t have a TikTok account, and can look in every direction at every moment. But if the best we can do is a 9x improvement of course we should still deploy it. The question is whether society will accept it. I’m not sure it would. I’m not sure it would accept a system that reduced deaths by two orders of magnitude, or 100x. Imagine for a moment, Tesla fully cracks the FSD nut, and every car on the road is replaced with a self driving Tesla, 100 times safer than humans. “One person is dying at the hands of the pedestrian butcher Elon Musk every single day!” the headlines will cry.

friend_and_foe 3y ago

While I agree with you, my hypothetical scenario above was deliberately constructed to avoid ascribing worth to distinct people, and only measuring how much of the outcome was the result of their negligent or deliberate behavior. We can come to the same conclusions while still assuming all lives have equal worth. All we have to do is determine where the fault lies. If a person jumps out in front of my car because they want to die, that's not my fault. If I run a stop sign and t bone a person it is. The same if I'm a computer in the trunk of a Tesla, and we want to be sure that if we are going to construct an environment where machines call the shots that they're less dangerous, so we should be sure that at the very least they're causing less accidents than people, not simply that they're participating in less accidents. Just the number of deaths doesn't give us this measure, it doesn't actually tell us how safe the machine is. What we ultimately want is primarily that the machine is less often at fault than a human driver, secondarily but very importantly that the machine avoids harming others even when those others are negligent and at fault, and also can avoid harming people in situations deliberately caused by others, particularly if those harmed weren't those causing the situation.