4 ms·

Trolley problem

22 points by networked 12y ago

plg 12y ago
I wonder how Tesla software will handle this problem. Also who will get sued when it actually happens.
- davidwihl 12y ago
  Like an autopilot in a plane, I imagine that there are some problems that a self driving car cannot or should not attempt to solve. This shifts the ethical and moral burden to the human driver as it should be. Though Armstrong in "Smarter Than Us: The Rise of Machine Intelligence" makes an interesting point: "these programs [autopilots and stock trading programs] occasionally encounter unexpected situations where humans must override, correct or rewrite them. But these overseers, who haven't been following the intricacies of the algorithm's decision process and who don't have the hands-on experience of the situation, are often at a complete loss as to what to do - and the plane or the stock market crashes." This is pretty much what happened in AF447 (http://en.wikipedia.org/wiki/Air_France_Flight_447 http://en.wikipedia.org/wiki/Air_France_Flight_447). In a self-driving car case, I hope the human will be able to exercise at least a defensible judgement. I find the future of ethics of intelligent machines fascinating. I plan to put a lot more energy into this area.
- ghaff 12y ago
  We're far far away from autonomous driving software being at a state where this is an interesting question. And, PR aside, Tesla's latest update is the same sort of assistive driving tech that's already in a number of high-end cars. It's not self-driving in that the driver is supposed to remain alert (and remains fully responsible for being in control of the vehicle). We do seem to be rapidly heading toward an "interesting" place where assistive driving technologies are good enough that a lot of people realistically won't pay attention while driving (given that plenty of people text and use their phones even without such technologies).
  - cjbprime 12y ago
    You seem overly pessimistic about autonomous driving software; it seems to me like Google's driving software could easily already be at a point where it might be able to consider the choice between hitting pedestrians crossing the street, or purposefully crashing the car.
    - ghaff 12y ago
      As a general approach, I can certainly see assistive driving algorithms making decisions about whether to veer off the road vs. possibly hitting something ahead, i.e. how do you prioritize staying in the lane vs. avoiding a known/likely collision. One implication is that you'd sort of like to know what "off the road" means in a given location.
      - walshemj 12y ago
        And what happens if you have an out of date map and the car thinks its swerving into an empty lot - but its now a kids playground.
    - brudgers 12y ago
      It's my understanding that Google's self-driving car technology as shown to the public, cannott currently handle a crew repairing potholes without advance notice, or drive to an arbitrary location using a route that hasn't been pre-run to record navigational waypoints. In other words, the disclosed technologies can drive in areas for which extremely large amounts of data have been collected and only so long as conditions have not changed. If a new traffic light or stop sign is installed, the self-driving car is likely to miss it. There may be secret sauce in the lab, of course.
  - s_kilk 12y ago
    It'll be interesting to see what happens the first time a self-driving car fucks up and ploughs through a bus queue. Who would be responsible? The "driver" who was chilling out reading their kindle? The manufacturer?
simonh 12y ago
Who needs karma anyway: http://xkcd.com/1455/ http://xkcd.com/1455/
kens 12y ago
This parody of the trolley problem is amusing: http://www.mindspring.com/~mfpatton/Tissues.htm http://www.mindspring.com/~mfpatton/Tissues.htm
zo1 12y ago
Has anyone considered the first option? I.e. not actually touching the lever. At least, in the human example of the problem, and not the one related to autonomous vehicles. If so, what do you think of it? I'm of the opinion that touching the lever would make you liable for the death/harm of the single individual. Sure, you "saved" 5 people and prevented them from coming to harm. But you knowingly decided to harm someone completely innocent in all of this. In my book, that's as good as letting go of the trolley in the first place. i.e. It's no longer an accident, but rather a deliberate (and arguably malicious) act.
- cjbprime 12y ago
  The Wikipedia article says that 90% choose to touch the lever, and 10% choose the first option. But people are more in agreement with you than it seems, because changing the formulation to the (ridiculously impossible) "fat man" version of the thought experiment causes most people to change their mind, even though the morally salient features of the experiment (choosing to kill one innocent person in order to save five innocent people) have not changed.
- eloisant 12y ago
  This is precisely what the "trolley problem" aims to highlight, and why the question is interesting rather than just being answered "obviously you'll choose one death rather than five" by everyone.
hurin 12y ago
A circle jerk of academics salaried for writing useless things about a hypothetical situation.
- m_y-n_a_m_e 12y ago
  "Sidetracked by trolleys: why sacrificial moral dilemmas tell us little (or nothing) about utilitarian judgment". Soc Neurosci. 2015 Mar 20:1-10. [Epub ahead of print] - http://www.ncbi.nlm.nih.gov/pubmed/25791902?dopt=Abstract http://www.ncbi.nlm.nih.gov/pubmed/25791902?dopt=Abstract tldr; ".... Sacrificial dilemmas therefore tell us little about utilitarian decision-making". Paper proposes to "studying proto-utilitarian tendencies in everyday moral thinking" - seems a good option to choose.
nakedrobot2 12y ago
KILL FEWER PEOPLE. What is the problem, here?
- gweinberg 12y ago
  Overpopulation?
brudgers 12y ago
The problem with the trolley problem is that it discounts time. In the time between throwing or not throwing the switch and the trolley reaching the location of the potential victim[s], our experience gives us some small hope that the victim[s] might be rescued by others. themselves, or trolley malfunction. There is still time for good fortune to replace the bad. The problem does not account for our ordinary intuition in this regard. This explains the apparent conundrum of divergence in the Fat Man version. If we push him off the bridge, there is no time for good fortune to save him, but there is time for the group to escape, self preserve or get lucky.
petersouth 12y ago
...only a real question if the identities of the people are known.

Trolley problem

22 points by networked 12y ago

plg 12y ago

I wonder how Tesla software will handle this problem. Also who will get sued when it actually happens.

davidwihl 12y ago
Like an autopilot in a plane, I imagine that there are some problems that a self driving car cannot or should not attempt to solve. This shifts the ethical and moral burden to the human driver as it should be. Though Armstrong in "Smarter Than Us: The Rise of Machine Intelligence" makes an interesting point: "these programs [autopilots and stock trading programs] occasionally encounter unexpected situations where humans must override, correct or rewrite them. But these overseers, who haven't been following the intricacies of the algorithm's decision process and who don't have the hands-on experience of the situation, are often at a complete loss as to what to do - and the plane or the stock market crashes." This is pretty much what happened in AF447 (http://en.wikipedia.org/wiki/Air_France_Flight_447 http://en.wikipedia.org/wiki/Air_France_Flight_447). In a self-driving car case, I hope the human will be able to exercise at least a defensible judgement. I find the future of ethics of intelligent machines fascinating. I plan to put a lot more energy into this area.
ghaff 12y ago
We're far far away from autonomous driving software being at a state where this is an interesting question. And, PR aside, Tesla's latest update is the same sort of assistive driving tech that's already in a number of high-end cars. It's not self-driving in that the driver is supposed to remain alert (and remains fully responsible for being in control of the vehicle). We do seem to be rapidly heading toward an "interesting" place where assistive driving technologies are good enough that a lot of people realistically won't pay attention while driving (given that plenty of people text and use their phones even without such technologies).
- cjbprime 12y ago
  You seem overly pessimistic about autonomous driving software; it seems to me like Google's driving software could easily already be at a point where it might be able to consider the choice between hitting pedestrians crossing the street, or purposefully crashing the car.
  - ghaff 12y ago
    As a general approach, I can certainly see assistive driving algorithms making decisions about whether to veer off the road vs. possibly hitting something ahead, i.e. how do you prioritize staying in the lane vs. avoiding a known/likely collision. One implication is that you'd sort of like to know what "off the road" means in a given location.
    - walshemj 12y ago
      And what happens if you have an out of date map and the car thinks its swerving into an empty lot - but its now a kids playground.
  - brudgers 12y ago
    It's my understanding that Google's self-driving car technology as shown to the public, cannott currently handle a crew repairing potholes without advance notice, or drive to an arbitrary location using a route that hasn't been pre-run to record navigational waypoints. In other words, the disclosed technologies can drive in areas for which extremely large amounts of data have been collected and only so long as conditions have not changed. If a new traffic light or stop sign is installed, the self-driving car is likely to miss it. There may be secret sauce in the lab, of course.
- s_kilk 12y ago
  It'll be interesting to see what happens the first time a self-driving car fucks up and ploughs through a bus queue. Who would be responsible? The "driver" who was chilling out reading their kindle? The manufacturer?

simonh 12y ago

Who needs karma anyway: http://xkcd.com/1455/ http://xkcd.com/1455/

kens 12y ago

This parody of the trolley problem is amusing: http://www.mindspring.com/~mfpatton/Tissues.htm http://www.mindspring.com/~mfpatton/Tissues.htm

zo1 12y ago

Has anyone considered the first option? I.e. not actually touching the lever. At least, in the human example of the problem, and not the one related to autonomous vehicles. If so, what do you think of it? I'm of the opinion that touching the lever would make you liable for the death/harm of the single individual. Sure, you "saved" 5 people and prevented them from coming to harm. But you knowingly decided to harm someone completely innocent in all of this. In my book, that's as good as letting go of the trolley in the first place. i.e. It's no longer an accident, but rather a deliberate (and arguably malicious) act.

cjbprime 12y ago
The Wikipedia article says that 90% choose to touch the lever, and 10% choose the first option. But people are more in agreement with you than it seems, because changing the formulation to the (ridiculously impossible) "fat man" version of the thought experiment causes most people to change their mind, even though the morally salient features of the experiment (choosing to kill one innocent person in order to save five innocent people) have not changed.
eloisant 12y ago
This is precisely what the "trolley problem" aims to highlight, and why the question is interesting rather than just being answered "obviously you'll choose one death rather than five" by everyone.

hurin 12y ago

A circle jerk of academics salaried for writing useless things about a hypothetical situation.

m_y-n_a_m_e 12y ago
"Sidetracked by trolleys: why sacrificial moral dilemmas tell us little (or nothing) about utilitarian judgment". Soc Neurosci. 2015 Mar 20:1-10. [Epub ahead of print] - http://www.ncbi.nlm.nih.gov/pubmed/25791902?dopt=Abstract http://www.ncbi.nlm.nih.gov/pubmed/25791902?dopt=Abstract tldr; ".... Sacrificial dilemmas therefore tell us little about utilitarian decision-making". Paper proposes to "studying proto-utilitarian tendencies in everyday moral thinking" - seems a good option to choose.

nakedrobot2 12y ago

KILL FEWER PEOPLE. What is the problem, here?

gweinberg 12y ago
Overpopulation?

brudgers 12y ago

The problem with the trolley problem is that it discounts time. In the time between throwing or not throwing the switch and the trolley reaching the location of the potential victim[s], our experience gives us some small hope that the victim[s] might be rescued by others. themselves, or trolley malfunction. There is still time for good fortune to replace the bad. The problem does not account for our ordinary intuition in this regard. This explains the apparent conundrum of divergence in the Fat Man version. If we push him off the bridge, there is no time for good fortune to save him, but there is time for the group to escape, self preserve or get lucky.

petersouth 12y ago

...only a real question if the identities of the people are known.