I would hope that the logic in saving the driver would ultimately save the pedestrian. I'd think that the "collision" of any object, including that of the body of a human being, could potentially do damage to the driver.
In extreme corners of the state space that doesn't hold, but one would hope self-driving cars would manage to stay far away from most of those corners.
For example, if a self-driving car is heading towards a concrete wall at speed, the safest thing to do is decelerate just fast enough to come to a stop touching or just after hitting the wall. If there are pedestrians between car and wall, Mercedes' statement seems to imply they will use the pedestrians as an additional crumple zone, rather than to decelerate really fast in order to not hit them, even if that saves lives and only causes the passengers some bruises.
More realistic: a pedestrian starts crossing the road between parked vans with darkened windows; a self-driving car doesn't see him until 10 meters away. Should the car swerve into the first van or into oncoming traffic to prevent the collision with a pedestrian, just brake as fast as possible, or brake as fast as is safe for the driver and the passengers? If the car knows the passengers did not put on their seat belts and are carrying loose heavy baggage on the rear shelf, should the car decide to decelerate very lightly, pedestrian be damned? What if this is in a pedestrian zone, and the driver overrode the car's intentions to keep speed below 10mph?
Another extreme example: a Mercedes is driving to pick up its owner, and has to decide between lightly harming its owner and killing a stranger. What will it do?
I doubt Mercedes' statement ("zeroeth law of robotics: A robot may not injure its owner or, through inaction, allow its owner to come to harm") will hold up in court, and also whether they will really implement it to the limit.
These make for interesting thought experiments, but how often will the observable data lead one to a clear cut reliable belief that action A causes X deaths and action B causes Y deaths? And it only gets more challenging if the injures are typically going to be non-life threatening in most cases but could still cause death (there are plenty of cases of people receiving a punch that normally would just bruise but thanks to a stumble or hitting a pavement edge they end up dead?)
That will change soon enough. The probability of surviving in a car crash is higher than getting run over by a car. These cars have seat belts, airbags, solid frames.
It would be interesting to see what algorithm they use to determine who gets saved. A lawsuit will change that fast enough.
I can already see a case, The speed limit is 45mph, someone overrides the car and forces it to go faster 70mph, something happens and the car has to make a decision, do you save the driver or the pedestrian who has the right of way?
Let's say Mercedes says the pedestrian, well, these are kids going to school, would they still save the driver?
Perhaps they would program the car to save the pedestrian if kids, else the driver. Maybe they would take into account the socioeconomic state of the environment. I firmly believe that a law would be passed in the future, where the pedestrian would be saved, but then what happens if an idiot jumps into the street/highway, should you die or have the car run them over? As we can see this is not simple, I think places are going to have laws and rules on who gets saved and how. In California the driver gets saved, in NY the pedestrian. It's going to be a mess.
> One of the biggest debates about driverless cars concerns the moral choices made when programming a car's algorithms.
This garbage should be downvoted on HN. Even on the subject of self-driving cars, these articles are a distraction, clickbait suitable for twitter arguments.
There is not much to discuss in a practical way here, because the fear of the incident related to Mercedes's statement is near totally superfluous.
Many are imagining a world similar to our own today, but with self-driving cars; this kind of effort and R&D is meant to be applied graceful over decades, which isn't what Mercedes is planning for today and shouldn't be contextualized within out current modernity.
What I believe is envisioned (obviously IMO) and why I feel this is a superfluous discussion is because in the world of self-driving cars of the Level 4 and up variety we are talking about beacon detection e.g. there are no pack of kids walking behind parked vans, because they all carry detectable devices such as laptops that always run Wifi, iPads, iPhones, etc. Everywhere they go these kids are detected and reacted in a world of Level 4+ cars. This is the same for pedestrians. Buses are also self-driving and all are communicating with each other in real-time and in fractions of a second.
A Mercedes car in this world didn't just discover a patch of ice on the road that went previously unnoticed. The previous 100+ cars that drove over it detected a level of decreased and decreasing level of traction (probably on a predictable curve I'd even assume) and communicated it to the network that coordinates cars, traffic, pedestrian signals. It's insane to think that in a world of Lidar systems, infrared, AI and so that these things will just 'happen' as they do today; that isn't realistic or even reasonable.
Ice? Oil spills, diesel spills, specs of glass, even a snot rocket from your local motorcycle gang are all detected immediately and compensated for with heat, humidity, tire traction quality deviation etc. It'll be just short of outright freaking amazing what this technology will eventually be able to do.
For me this is just fun conversation, but ultimately totally useless beyond the ethical dilemma (which is more of a Pandora box discussion than anything), because it just won't really exist in any material manner in the future. Like a chicken or the egg. ;)
Sorry haven't had a reason to use that word in a long long time; too much fun to pass on.
8 comments
[ 1.6 ms ] story [ 29.0 ms ] threadFor example, if a self-driving car is heading towards a concrete wall at speed, the safest thing to do is decelerate just fast enough to come to a stop touching or just after hitting the wall. If there are pedestrians between car and wall, Mercedes' statement seems to imply they will use the pedestrians as an additional crumple zone, rather than to decelerate really fast in order to not hit them, even if that saves lives and only causes the passengers some bruises.
More realistic: a pedestrian starts crossing the road between parked vans with darkened windows; a self-driving car doesn't see him until 10 meters away. Should the car swerve into the first van or into oncoming traffic to prevent the collision with a pedestrian, just brake as fast as possible, or brake as fast as is safe for the driver and the passengers? If the car knows the passengers did not put on their seat belts and are carrying loose heavy baggage on the rear shelf, should the car decide to decelerate very lightly, pedestrian be damned? What if this is in a pedestrian zone, and the driver overrode the car's intentions to keep speed below 10mph?
Another extreme example: a Mercedes is driving to pick up its owner, and has to decide between lightly harming its owner and killing a stranger. What will it do?
I doubt Mercedes' statement ("zeroeth law of robotics: A robot may not injure its owner or, through inaction, allow its owner to come to harm") will hold up in court, and also whether they will really implement it to the limit.
It would be interesting to see what algorithm they use to determine who gets saved. A lawsuit will change that fast enough.
I can already see a case, The speed limit is 45mph, someone overrides the car and forces it to go faster 70mph, something happens and the car has to make a decision, do you save the driver or the pedestrian who has the right of way?
Let's say Mercedes says the pedestrian, well, these are kids going to school, would they still save the driver?
Perhaps they would program the car to save the pedestrian if kids, else the driver. Maybe they would take into account the socioeconomic state of the environment. I firmly believe that a law would be passed in the future, where the pedestrian would be saved, but then what happens if an idiot jumps into the street/highway, should you die or have the car run them over? As we can see this is not simple, I think places are going to have laws and rules on who gets saved and how. In California the driver gets saved, in NY the pedestrian. It's going to be a mess.
This garbage should be downvoted on HN. Even on the subject of self-driving cars, these articles are a distraction, clickbait suitable for twitter arguments.
Many are imagining a world similar to our own today, but with self-driving cars; this kind of effort and R&D is meant to be applied graceful over decades, which isn't what Mercedes is planning for today and shouldn't be contextualized within out current modernity.
What I believe is envisioned (obviously IMO) and why I feel this is a superfluous discussion is because in the world of self-driving cars of the Level 4 and up variety we are talking about beacon detection e.g. there are no pack of kids walking behind parked vans, because they all carry detectable devices such as laptops that always run Wifi, iPads, iPhones, etc. Everywhere they go these kids are detected and reacted in a world of Level 4+ cars. This is the same for pedestrians. Buses are also self-driving and all are communicating with each other in real-time and in fractions of a second.
A Mercedes car in this world didn't just discover a patch of ice on the road that went previously unnoticed. The previous 100+ cars that drove over it detected a level of decreased and decreasing level of traction (probably on a predictable curve I'd even assume) and communicated it to the network that coordinates cars, traffic, pedestrian signals. It's insane to think that in a world of Lidar systems, infrared, AI and so that these things will just 'happen' as they do today; that isn't realistic or even reasonable.
Ice? Oil spills, diesel spills, specs of glass, even a snot rocket from your local motorcycle gang are all detected immediately and compensated for with heat, humidity, tire traction quality deviation etc. It'll be just short of outright freaking amazing what this technology will eventually be able to do.
For me this is just fun conversation, but ultimately totally useless beyond the ethical dilemma (which is more of a Pandora box discussion than anything), because it just won't really exist in any material manner in the future. Like a chicken or the egg. ;)
Sorry haven't had a reason to use that word in a long long time; too much fun to pass on.