3 ms·

Have you worked in AV? “Save Driver” vs “Save Pedestrian” isn’t actually a problem that comes up in the real world. There’s never a situation where you’d use a

by ponker 6y ago

Have you worked in AV? “Save Driver” vs “Save Pedestrian” isn’t actually a problem that comes up in the real world. There’s never a situation where you’d use a pedestrian as a meat bag to absorb a collision or fling a driver over a median to save a pedestrian. The dilemma is completely a philosophical construct, no self driving car will ever need to solve it. In reality, saving drivers and saving pedestrians are completely orthogonal problems.

luckylion 6y ago
That's not the typical problem in AV's trolley problem, nobody argues (publicly) "the driver should always be first". You don't even need the driver to play a role, you can use an AV truck, train or drone. "Your brakes are gone, you cannot decrease speed. You are on a collision course with 10 people who are crossing the street. You can slightly alter your direction to run over one person instead." You can alter that as required, put a dog in the other lane. Or an old person. If you want to go deeper: put a light car in the other lane with the assumption that the driver of that car has a 10% chance to survive. Give http://moralmachine.mit.edu/ http://moralmachine.mit.edu/ a try, they've set up a few different scenarios. "None of this will ever happen, the car will simply learn to fly and avoid the accident" is not an option unfortunately.
- ponker 6y ago
  What I am saying is that in for both meatbag and AI drivers, the “brakes failed headed towards a crowd” situation is so rare that the driver has neither the time nor experience/training data to make any kind of well-reasoned ethical decisions, and basically ends up performing an action that is close to random, and that the situation is so rare that the impact isn’t relevant when policymaking/regulating AV. It’s like the difference between the deaths from drunk drivers and the deaths from drivers being stung by bees (which do exist, but not enough to make rules like every car needs a bee repellent).
  - luckylion 6y ago
    Essentially: it's rare, we shouldn't worry about it? I don't think that's a valid argument. A driver may not have a lot of time to ponder the consequences of his actions (but you can create a hypothetical that would give him a few seconds, of course, but hypothetical situations are an issue in this thread), but that doesn't mean the AV that's not as limited in recognition and calculations shouldn't calculate probable outcomes and make a decision, even if it's rare. And once you agree that it should if it can, there's no sense in saying "but we shouldn't talk about what it should optimize for". And of course, we can generally expect that if it is possible, that it will be done and Mercedes will ask: who's paying us, the person in the car or the person outside the car, so they will gravitate towards saving their customer. At that point you have made the decision, but you've avoided having a debate over ethics.
    - ponker 6y ago
      Essentially: the situation only exists in the contrived world of ethicists, so the problem may be important for ethicists to solve but isn't important for AV engineers to solve. It's like asking a parent the question, "If both your kids were falling off a cliff at the exact same time and you could only save one of them, which one would you save." The parents says "well whichever one is closer to me." The ethicist says "No they are the exact same distance from you, with the one near your non-dominant arm exactly the amount closer to you to cancel out the advantage of your dominant arm." Well, Okay. It's a stressful question to be asked and to try to answer, but the answer in the actual real world is that the situation doesn't actually exist, and even if you were to replay the universe 5 quintillion times until the situation came up, the flesh and blood parent would panic in the moment and save neither one. So ethicists can debate the situation ad infinitum and entertain themselves doing so, but the situation only exists in the contrived world, so the answer doesn't matter.
      - luckylion 6y ago
        You absolutely cannot imagine a situation where a car will most likely cause damage but the driver (be that a computer or a human) can influence where it hits? And if it can be influenced, why wouldn't you want the least amount of damage vs some random amount of damage? I'm not talking about Descartes and his all-powerful god that may fool him. "Which of your kids do you send to college if you don't have enough money to send both?" is a similar question, and that may be unthinkable and totally unrealistic for a SWE, but it's a pretty normal situation for lots of people.

by ponker 6y ago

luckylion 6y ago

That's not the typical problem in AV's trolley problem, nobody argues (publicly) "the driver should always be first". You don't even need the driver to play a role, you can use an AV truck, train or drone. "Your brakes are gone, you cannot decrease speed. You are on a collision course with 10 people who are crossing the street. You can slightly alter your direction to run over one person instead." You can alter that as required, put a dog in the other lane. Or an old person. If you want to go deeper: put a light car in the other lane with the assumption that the driver of that car has a 10% chance to survive. Give http://moralmachine.mit.edu/ http://moralmachine.mit.edu/ a try, they've set up a few different scenarios. "None of this will ever happen, the car will simply learn to fly and avoid the accident" is not an option unfortunately.

ponker 6y ago
What I am saying is that in for both meatbag and AI drivers, the “brakes failed headed towards a crowd” situation is so rare that the driver has neither the time nor experience/training data to make any kind of well-reasoned ethical decisions, and basically ends up performing an action that is close to random, and that the situation is so rare that the impact isn’t relevant when policymaking/regulating AV. It’s like the difference between the deaths from drunk drivers and the deaths from drivers being stung by bees (which do exist, but not enough to make rules like every car needs a bee repellent).
- luckylion 6y ago
  Essentially: it's rare, we shouldn't worry about it? I don't think that's a valid argument. A driver may not have a lot of time to ponder the consequences of his actions (but you can create a hypothetical that would give him a few seconds, of course, but hypothetical situations are an issue in this thread), but that doesn't mean the AV that's not as limited in recognition and calculations shouldn't calculate probable outcomes and make a decision, even if it's rare. And once you agree that it should if it can, there's no sense in saying "but we shouldn't talk about what it should optimize for". And of course, we can generally expect that if it is possible, that it will be done and Mercedes will ask: who's paying us, the person in the car or the person outside the car, so they will gravitate towards saving their customer. At that point you have made the decision, but you've avoided having a debate over ethics.
  - ponker 6y ago
    Essentially: the situation only exists in the contrived world of ethicists, so the problem may be important for ethicists to solve but isn't important for AV engineers to solve. It's like asking a parent the question, "If both your kids were falling off a cliff at the exact same time and you could only save one of them, which one would you save." The parents says "well whichever one is closer to me." The ethicist says "No they are the exact same distance from you, with the one near your non-dominant arm exactly the amount closer to you to cancel out the advantage of your dominant arm." Well, Okay. It's a stressful question to be asked and to try to answer, but the answer in the actual real world is that the situation doesn't actually exist, and even if you were to replay the universe 5 quintillion times until the situation came up, the flesh and blood parent would panic in the moment and save neither one. So ethicists can debate the situation ad infinitum and entertain themselves doing so, but the situation only exists in the contrived world, so the answer doesn't matter.
    - luckylion 6y ago
      You absolutely cannot imagine a situation where a car will most likely cause damage but the driver (be that a computer or a human) can influence where it hits? And if it can be influenced, why wouldn't you want the least amount of damage vs some random amount of damage? I'm not talking about Descartes and his all-powerful god that may fool him. "Which of your kids do you send to college if you don't have enough money to send both?" is a similar question, and that may be unthinkable and totally unrealistic for a SWE, but it's a pretty normal situation for lots of people.