|
How often have people crossed the street where they didn't calculate howmuch time they had to cross it?
I feel like the self driving car software you purchase for your car is going to be in classes, where the cheapest isn't really gonna give a shit about you and the most expensive is going to save you at all cost.
"Would you like the Kamikaze package or the People Mower package?"
|
On October 13 2016 02:13 Uldridge wrote:
How often have people crossed the street where they didn't calculate howmuch time they had to cross it?
I feel like the self driving car software you purchase for your car is going to be in classes, where the cheapest isn't really gonna give a shit about you and the most expensive is going to save you at all cost.
"Would you like the Kamikaze package or the People Mower package?"
I'd like the mild mower with slight chance of kamikaze for heroic encounters. Just enough kamikazi for honor, but not enough for danger.
|
On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen?
Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it.
And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings.
edit : and yeah I'm sorry but before we look into the morals of driverless cars, I'd like it if people around the globe could look into the morals of drivers who create accident-prone situations literally everytime they get into their car and don't even realize it.
|
On October 13 2016 02:41 OtherWorld wrote:Show nested quote +On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings.
So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused.
|
On October 13 2016 02:46 Thieving Magpie wrote:Show nested quote +On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused.
And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them.
e : I'm actually pretty sure you can create accidents without much physical risk for yourself with regular drivers in the cars. Why don't we see anachists trying to wreck havoc everywhere, then?
|
On October 13 2016 02:48 OtherWorld wrote:Show nested quote +On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them.
Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be.
|
On October 13 2016 02:54 Thieving Magpie wrote:Show nested quote +On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be.
You don't. AI learn over time, they're, precisely, (supposedly) intelligent. Besides, you can do that right now, without driverless cars : get your gang, take 1 or 2 of your rivals, throw them into the street too late for drivers to brake. Still works : either your rivals die/suffer heavy injuries, or you created an accident involving several cars (or you got the combo). Why isn't this already a mass problem, then?
This is even truer as driverless cars can brake before driverful cars can, and can also brake harder than drivers can, and thus will be carrying less kinetic energy onto whatever they hit, diminishing the risk for heavy injury or death.
|
On October 13 2016 02:54 Thieving Magpie wrote:Show nested quote +On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be.
And what advantage would abusing the AI of an automated car to kill someone serve over killing them using the plethora of normal means already available?
Besides an automated car has much better "reflexes" than a human being, so trying to manufacture a scenario that guarantees that the AI will choose to kill someone isn't reasonable.
|
On October 13 2016 03:04 ZigguratOfUr wrote:Show nested quote +On October 13 2016 02:54 Thieving Magpie wrote:On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be. And what advantage would abusing the AI of an automated car to kill someone serve over killing them using the plethora of normal means already available? Besides an automated car has much better "reflexes" than a human being, so trying to manufacture a scenario that guarantees that the AI will choose to kill someone isn't reasonable.
You're right, why would I think a video game forum does not believe its believable to abuse AI.
|
On October 13 2016 03:13 Thieving Magpie wrote:Show nested quote +On October 13 2016 03:04 ZigguratOfUr wrote:On October 13 2016 02:54 Thieving Magpie wrote:On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be. And what advantage would abusing the AI of an automated car to kill someone serve over killing them using the plethora of normal means already available? Besides an automated car has much better "reflexes" than a human being, so trying to manufacture a scenario that guarantees that the AI will choose to kill someone isn't reasonable. You're right, why would I think a video game forum does not believe its believable to abuse AI.
It is possible to abuse AI, and nobody denied that. You didn't answer the points, though :
-Why would abusing the AI be an issue while abusing actual humans apparently isn't?
-Why would an AI be held responsible for killing someone after a split-second decision, while a human driver doing the same apparently isn't responsible?
-Since AI cars are faster to react than humans, why would an AI car be more dangerous?
But no, much better to resort to ad hominems.
|
On October 13 2016 03:17 OtherWorld wrote:Show nested quote +On October 13 2016 03:13 Thieving Magpie wrote:On October 13 2016 03:04 ZigguratOfUr wrote:On October 13 2016 02:54 Thieving Magpie wrote:On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be. And what advantage would abusing the AI of an automated car to kill someone serve over killing them using the plethora of normal means already available? Besides an automated car has much better "reflexes" than a human being, so trying to manufacture a scenario that guarantees that the AI will choose to kill someone isn't reasonable. You're right, why would I think a video game forum does not believe its believable to abuse AI. It is possible to abuse AI, and nobody denied that. You didn't answer the points, though : -Why would abusing the AI be an issue while abusing actual humans apparently isn't? -Why would an AI be held responsible for killing someone after a split-second decision, while a human driver doing the same apparently isn't responsible? -Since AI cars are faster to react than humans, why would an AI car be more dangerous? But no, much better to resort to ad hominems.
I already made that argument. You know what AI will do, you won't know what people will do. People might kill them, they might swerve into the sidewalk and kill the guy pushing, they might swerve into traffic and hurt everyone BUT the guy pushed.
And it doesn't have to be planned. What if an altercation happened while out drinking. During the heat of the fight you see a car, and you already know there's no way that car would do X, Y, and Z because programming already tells it to do A, B, and C instead. So the likelihood of being willing to use that as a tool becomes more of an option as opposed to today where the randomness of human response would be bad.
It could even be knowing that cars cannot do that and so its no longer an option. Imagine if you knew cars would not run into you if you walked into the street, so you do what lots of people in my city does now and just cross traffic whenever. You know the chances of the driver not seeing you is almost zero so you just walk across the street whenever and wherever. Why not? If you know it will guarantee kill someone then you use it for that instead. People will find ways to flow around any automated system. Being overly nitpicky about a specific way to leverage that system misses the whole point.
|
On October 13 2016 03:25 Thieving Magpie wrote:Show nested quote +On October 13 2016 03:17 OtherWorld wrote:On October 13 2016 03:13 Thieving Magpie wrote:On October 13 2016 03:04 ZigguratOfUr wrote:On October 13 2016 02:54 Thieving Magpie wrote:On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be. And what advantage would abusing the AI of an automated car to kill someone serve over killing them using the plethora of normal means already available? Besides an automated car has much better "reflexes" than a human being, so trying to manufacture a scenario that guarantees that the AI will choose to kill someone isn't reasonable. You're right, why would I think a video game forum does not believe its believable to abuse AI. It is possible to abuse AI, and nobody denied that. You didn't answer the points, though : -Why would abusing the AI be an issue while abusing actual humans apparently isn't? -Why would an AI be held responsible for killing someone after a split-second decision, while a human driver doing the same apparently isn't responsible? -Since AI cars are faster to react than humans, why would an AI car be more dangerous? But no, much better to resort to ad hominems. I already made that argument. You know what AI will do, you won't know what people will do. People might kill them, they might swerve into the sidewalk and kill the guy pushing, they might swerve into traffic and hurt everyone BUT the guy pushed. And it doesn't have to be planned. What if an altercation happened while out drinking. During the heat of the fight you see a car, and you already know there's no way that car would do X, Y, and Z because programming already tells it to do A, B, and C instead. So the likelihood of being willing to use that as a tool becomes more of an option as opposed to today where the randomness of human response would be bad. It could even be knowing that cars cannot do that and so its no longer an option. Imagine if you knew cars would not run into you if you walked into the street, so you do what lots of people in my city does now and just cross traffic whenever. You know the chances of the driver not seeing you is almost zero so you just walk across the street whenever and wherever. Why not? If you know it will guarantee kill someone then you use it for that instead. People will find ways to flow around any automated system. Being overly nitpicky about a specific way to leverage that system misses the whole point.
You know what else is guaranteed to kill someone? A bullet. Much more reliable than trying to figure out what an AI will do when faced with a specific scenario (accident prevention and damage mitigation software won't just be a flowchart), and trying to create that scenario. Self-driven cars won't make the result of pushing someone into traffic any more obvious or lethal than currently.
I'm sure people will try to leverage their knowledge of car AIs, just as they already try to leverage their knowledge of people or anything else. I'm sure some people will be moderately successful in doing so. But claiming that mayhem must result from those "abuses" is specious.
|
On October 13 2016 03:46 ZigguratOfUr wrote:Show nested quote +On October 13 2016 03:25 Thieving Magpie wrote:On October 13 2016 03:17 OtherWorld wrote:On October 13 2016 03:13 Thieving Magpie wrote:On October 13 2016 03:04 ZigguratOfUr wrote:On October 13 2016 02:54 Thieving Magpie wrote:On October 13 2016 02:48 OtherWorld wrote:On October 13 2016 02:46 Thieving Magpie wrote:On October 13 2016 02:41 OtherWorld wrote:On October 13 2016 01:57 KwarK wrote:
It's much more complex than that. A driver in that situation has to do what he can to make the best splitsecond judgement and we don't second guess them too much for it. If he swerved to avoid an oncoming vehicle that would have killed him and hit a pedestrian then that's unfortunate but understandable as an instinctive response.
With self driving cars it's a completely different situation. There needs to be a system created that values different scenarios and decides upon the correct action and in doing so that system accepts liability for the outcome. If you have a tenth of a second to decide to hit a child or their grandpa you're not going to get judged, that's not even enough time to consider the choice. But the AI in the car isn't reacting on instinct, it's following programming coded by humans that had plenty of time to consider exactly how to resolve this situation. The programmers have to make a conscious choice, telling it which one to hit.
There are various different moral strategies the design could employ, from maximum good to protect the driver at all costs and everything in between. But in any of them you remove the element of accident from it, there is a conscious choice of who lives and who dies being made by the people creating the strategy, and that's very difficult terrain.
If there were two cars for sale with different self driving programming, one that was programmed to kill you if it was the least bad option, one programmed to always try to save you, which one would you buy? And if you bought the one that saved you are you accepting responsibility for the potential to mow down a family because they're softer than a wall, should that happen? Yet there is no "correct action" based on anything else than morals. And different people having different morals, I fail to see how anything productive can come out of it. And yeah, no, unless we under some kind of Watchdogs shit, the AI doesn't know if the guy it's about to hit is 80 or 8, or if he's a criminal or a doctor, or if he wants to suicide or is just a drunken twat. The only thing the AI could consider (and I don't even know if that's possible) is which action dissipates the most kinetic energy on something else than human beings. So many pranksters and anarchists will then start running into street to cause crashes collisions and mayhem. Automated systems serves nothing but to be abused. And what guarantee would they have that they wouldn't be the one to die? Because, I mean, they could also do the same thing without driverless cars if they had the guarantee that drivers would avoid them. Drivers have no guarantees. Automated drivers do. Like, imagine if it was guaranteed to run over the person--gangbangers would just throw 1-2 of their rivals into the street knowing the cars won't slow down. Or if its guaranteed to swerve, you can do the same thing. When its a human you'll never know what that person will do. Human could just as easily panic and swerve into the gangbanger if a person was thrown in front of them. But with AI you already know what their responses will be. And what advantage would abusing the AI of an automated car to kill someone serve over killing them using the plethora of normal means already available? Besides an automated car has much better "reflexes" than a human being, so trying to manufacture a scenario that guarantees that the AI will choose to kill someone isn't reasonable. You're right, why would I think a video game forum does not believe its believable to abuse AI. It is possible to abuse AI, and nobody denied that. You didn't answer the points, though : -Why would abusing the AI be an issue while abusing actual humans apparently isn't? -Why would an AI be held responsible for killing someone after a split-second decision, while a human driver doing the same apparently isn't responsible? -Since AI cars are faster to react than humans, why would an AI car be more dangerous? But no, much better to resort to ad hominems. I already made that argument. You know what AI will do, you won't know what people will do. People might kill them, they might swerve into the sidewalk and kill the guy pushing, they might swerve into traffic and hurt everyone BUT the guy pushed. And it doesn't have to be planned. What if an altercation happened while out drinking. During the heat of the fight you see a car, and you already know there's no way that car would do X, Y, and Z because programming already tells it to do A, B, and C instead. So the likelihood of being willing to use that as a tool becomes more of an option as opposed to today where the randomness of human response would be bad. It could even be knowing that cars cannot do that and so its no longer an option. Imagine if you knew cars would not run into you if you walked into the street, so you do what lots of people in my city does now and just cross traffic whenever. You know the chances of the driver not seeing you is almost zero so you just walk across the street whenever and wherever. Why not? If you know it will guarantee kill someone then you use it for that instead. People will find ways to flow around any automated system. Being overly nitpicky about a specific way to leverage that system misses the whole point. You know what else is guaranteed to kill someone? A bullet. Much more reliable than trying to figure out what an AI will do when faced with a specific scenario (accident prevention and damage mitigation software won't just be a flowchart), and trying to create that scenario. Self-driven cars won't make the result of pushing someone into traffic any more obvious or lethal than currently. I'm sure people will try to leverage their knowledge of car AIs, just as they already try to leverage their knowledge of people or anything else. I'm sure some people will be moderately successful in doing so. But claiming that mayhem must result from those "abuses" is specious.
You're right, what could possibly go wrong.
|
United States43991 Posts
I think people trying to metagame the AI is by far the least of the ethical concerns with it.
|
On October 13 2016 03:56 KwarK wrote:
I think people trying to metagame the AI is by far the least of the ethical concerns with it.
Agreed.
AIs will have to make life or death decisions where there isn't a correct answer (the I, Robot movie scenario basically). Any set of values that the programmer gives to the AI to make those decisions will carry its share of bias. And yet letting 98 people die because you don't want an AI to decide whether to save the 99th or the 100th is of course unthinkable.
|
On October 13 2016 04:04 ZigguratOfUr wrote:Show nested quote +On October 13 2016 03:56 KwarK wrote:
I think people trying to metagame the AI is by far the least of the ethical concerns with it. Agreed. AIs will have to make life or death decisions where there isn't a correct answer (the I, Robot movie scenario basically). Any set of values that the programmer gives to the AI to make those decisions will carry its share of bias. And yet letting 98 people die because you don't want an AI to decide whether to save the 99th or the 100th is of course unthinkable.
Or they'll just slowly transition to separate roads for AI and separate roads for Drivers and treat automated vehicles the same as trains today which are all fairly automated already.
|
Let's use even more infrastructure for the mass transit of people!
Let's make those people then live underground, or better yet, let's make it so there's one huge moving city on one of those AI roads and you can just plug in whenever you want.
Nothing beats traveling by city!
|
On October 13 2016 05:02 Uldridge wrote:
Let's use even more infrastructure for the mass transit of people!
Let's make those people then live underground, or better yet, let's make it so there's one huge moving city on one of those AI roads and you can just plug in whenever you want.
Nothing beats traveling by city!
What infrastructure? Just use the same ones we have now...
Examples:
AI lane, Norm Lane, Bike lane.
OR
"AI Road from X o'clock to Y o'clock, normal road from Z o'clock to X o'clock"
OR
"AI allowed on these neighborhoods, not these neighborhoods"
etc...
|
On October 13 2016 04:43 Thieving Magpie wrote:Show nested quote +On October 13 2016 04:04 ZigguratOfUr wrote:On October 13 2016 03:56 KwarK wrote:
I think people trying to metagame the AI is by far the least of the ethical concerns with it. Agreed. AIs will have to make life or death decisions where there isn't a correct answer (the I, Robot movie scenario basically). Any set of values that the programmer gives to the AI to make those decisions will carry its share of bias. And yet letting 98 people die because you don't want an AI to decide whether to save the 99th or the 100th is of course unthinkable. Or they'll just slowly transition to separate roads for AI and separate roads for Drivers and treat automated vehicles the same as trains today which are all fairly automated already.
The main problem with this is that there isn't a separate set of roads available for use for self-driving cars. In fact, the largest appeal of self-driving cars is that they could use the roads that are already there. If you are going to create a new set of roads, some sort of tracks would probably be better than roads.
To me, the main question for self-driving cars is "Are they safer than humans?". As far as i know, they already are. Yes, AI cars will produce some deaths. That is inevitable. But so do human-driven cars. The measure should not be "absolutely safe", it should be "Safer then the status quo". Yes, the exact details for the programming need some thought, and laws need to be changed to fit driverless cars. But that is not the big question. The big question is if you would want driverless cars. After you have figured that out, you need to think about details. But details shouldn't stand in the way of the big question.
In my opinion, once driverless cars become widespread, the days of human-driven cars are limited. There is simply not a lot that speaks for a human driver over a machine, except for tradition. They are less safe, and you waste multiple hours every day just sitting in front of the wheel steering a car, which you could otherwise use for productive tasks or leisure.
Also, driving is stressful and makes people angry.
|
Yeah, cars really are almost all bad. The only upside to cars is that they're reasonably fast.
|
|
|
|
|
|
2278
1079
260
188
4
EPT
