5 ms·

There's a (famous?) quote from a google engineer Andrew Chatham who worked on self driving cars. When asked about the trolley problem, he said "It takes some of

There's a (famous?) quote from a google engineer Andrew Chatham who worked on self driving cars. When asked about the trolley problem, he said "It takes some of the intellectual intrigue out of the problem, but the answer is almost always ‘slam on the brakes’” [0] [0]: https://www.theguardian.com/technology/2016/aug/22/self-driving-cars-moral-dilemmas https://www.theguardian.com/technology/2016/aug/22/self-driv...

synu 4y ago
You could adjust the question to be “the brakes have failed and..” to avoid that out.
- salford 4y ago
  True, I imagine the next course of action would be something like "do nothing" — continue straight, try to be predictable, hope that other people in the situation can react accordingly.
  - synu 4y ago
    If we are still doing the trolley problem it would be “the brakes have failed, and … do you continue straight and definitely hit A, or turn the wheel and definitely hit B”.
    - Dylan16807 4y ago
      Sure, that still works as a purely abstract problem. But when it comes to self-driving cars, the answer is that it doesn't really matter.
    - lynx09 4y ago
      You should not change lanes unless it is safe to do so. The car should let off the gas, put on hazards and try to move to the side of the road. I am with the other engineers, the trolley problem does not apply here. Why waste effort deciding who lives and who dies when more time can be spent figuring out how to handle these situations safely?
- dorgo 4y ago
  I know, you try to save the trolley scenario. But if brakes fail then it doesn't matter what the AI (or driver) decides. To blame is the one who is responsible for the brakes not to fail.
  - synu 4y ago
    Is the trolley scenario intended as a blame-finding simulation? I thought it was more like “what (ethical) choice do you make in a difficult situation” deal. You surely still have some agency after your brakes failed. Or at least if it happened to me I wouldn’t let go of the wheel to find out if I hit group A or B, confident in the knowledge that who or whatever I plow into it is not my fault and all choices from that point are equal.
  - galdosdi 4y ago
    What? There are still a lot of tactics a driver (or AI) can follow to improve the outcome if the brakes fail, such as swerving to avoid dangerous things while the vehicle naturally coasts to a stop, intentionally driving into a soft thing or up a hill if available (eg runaway truck ramp or a pile of haybales) or in the worst case, decide which of two bad things to run into is less bad. Not making a choice is a choice too, as in the trolley problem, it means you'll hit whatever is straight in front. So in that sense it's quite relevant. Great, you can usually just apply the brakes. What if you can't or don't have enough following distance? (less common for an AI driver that will try to leave enough distance, but that's not entirely in your own control; another driver can merge in front of you at any time they wish and if an incident occurs before you can increase the distance, you will effectively have to swerve since you won't have enough braking capacity. It is worthwhile to take a look at dashcam videos (eg, youtube, reddit) to get an idea of the many, many complex scenarios a driver can find themselves in in which a crash is imminent, but the driver has many options and must decide what strategy to take to get the least bad outcome.
  - dr_orpheus 4y ago
    But there are still plenty of either/or scenarios with a perfectly functional vehicle because there are other people on the road. Some of the interesting ones involve whether it is the AI car's responsibility to the drivers/passengers in the car first versus everyone else. For example, a car swerves in front of you (or runs a red light, or whatever) and the car hits the brakes but knows it will not be enough to stop running in to the car. Do they swerve off the road where they might kill a pedestrian, but will definitely save the driver or do you continue to crash in to the other car which may kill the driver of one or both cars.

synu 4y ago

You could adjust the question to be “the brakes have failed and..” to avoid that out.

salford 4y ago
True, I imagine the next course of action would be something like "do nothing" — continue straight, try to be predictable, hope that other people in the situation can react accordingly.
- synu 4y ago
  If we are still doing the trolley problem it would be “the brakes have failed, and … do you continue straight and definitely hit A, or turn the wheel and definitely hit B”.
  - Dylan16807 4y ago
    Sure, that still works as a purely abstract problem. But when it comes to self-driving cars, the answer is that it doesn't really matter.
  - lynx09 4y ago
    You should not change lanes unless it is safe to do so. The car should let off the gas, put on hazards and try to move to the side of the road. I am with the other engineers, the trolley problem does not apply here. Why waste effort deciding who lives and who dies when more time can be spent figuring out how to handle these situations safely?
dorgo 4y ago
I know, you try to save the trolley scenario. But if brakes fail then it doesn't matter what the AI (or driver) decides. To blame is the one who is responsible for the brakes not to fail.
- synu 4y ago
  Is the trolley scenario intended as a blame-finding simulation? I thought it was more like “what (ethical) choice do you make in a difficult situation” deal. You surely still have some agency after your brakes failed. Or at least if it happened to me I wouldn’t let go of the wheel to find out if I hit group A or B, confident in the knowledge that who or whatever I plow into it is not my fault and all choices from that point are equal.
- galdosdi 4y ago
  What? There are still a lot of tactics a driver (or AI) can follow to improve the outcome if the brakes fail, such as swerving to avoid dangerous things while the vehicle naturally coasts to a stop, intentionally driving into a soft thing or up a hill if available (eg runaway truck ramp or a pile of haybales) or in the worst case, decide which of two bad things to run into is less bad. Not making a choice is a choice too, as in the trolley problem, it means you'll hit whatever is straight in front. So in that sense it's quite relevant. Great, you can usually just apply the brakes. What if you can't or don't have enough following distance? (less common for an AI driver that will try to leave enough distance, but that's not entirely in your own control; another driver can merge in front of you at any time they wish and if an incident occurs before you can increase the distance, you will effectively have to swerve since you won't have enough braking capacity. It is worthwhile to take a look at dashcam videos (eg, youtube, reddit) to get an idea of the many, many complex scenarios a driver can find themselves in in which a crash is imminent, but the driver has many options and must decide what strategy to take to get the least bad outcome.
- dr_orpheus 4y ago
  But there are still plenty of either/or scenarios with a perfectly functional vehicle because there are other people on the road. Some of the interesting ones involve whether it is the AI car's responsibility to the drivers/passengers in the car first versus everyone else. For example, a car swerves in front of you (or runs a red light, or whatever) and the car hits the brakes but knows it will not be enough to stop running in to the car. Do they swerve off the road where they might kill a pedestrian, but will definitely save the driver or do you continue to crash in to the other car which may kill the driver of one or both cars.