3 ms·

I agree. But suppose the difference mentioned (1 sec) is true. That's such a large difference that what they call for (formal outside verification) sounds prett

by wjnc 9y ago

I agree. But suppose the difference mentioned (1 sec) is true. That's such a large difference that what they call for (formal outside verification) sounds pretty reasonable. This debate will always be hard since it will involve deaths, with AI and video involved.

reificator 9y ago
You're right. Developing video games has taught me that a second is a long time. But we still don't even know that Uber didn't detect them that early. The CV system or other sensors could have picked out the pedestrian but something else went wrong instead. I tend to think that if Uber did detect the pedestrian, they would have said something by now. But perhaps they prefer the optics of not having seen her to the optics of seeing her but not braking at all. We simply don't know. Imagine that press conference. > Uber: "After examining the black box on the car we have determined that the car detected the pedestrian but failed to brake." > Reporter: "Was this because the pedestrian was a woman?" > Reporter: "Did the car do this to protect the life of the Uber employee inside?" > Reporter: "Was her credit score too low?"
- 2474 9y ago
  This reminds me of a survey I had taken. I can't remember the survey publisher but it was related to self-driving vehicles. You were presented two pictures (e.g. a picture of school kids crossing a crosswalk and the other a picture of an adult male and female walking their dog across a crosswalk). Presumably you had to choose; would you swerve to hit the adults and dog or would you stay course and hit the kids. I know there is no moral compass for a computer system to determine what it should do in the scenario but I laughed at the quiz because I couldn't believe this is the kind of data that will feed machine-learning algorithms. It just feels bad.
  - lowbloodsugar 9y ago
    Getting data on an ethical dilemma. [1] [1] https://en.wikipedia.org/wiki/Trolley_problem https://en.wikipedia.org/wiki/Trolley_problem
    - 2474 9y ago
      Thanks for mentioning what that dilemma was called. For those interested here is an interesting article about the Trolley Problem. https://www.currentaffairs.org/2017/11/the-trolley-problem-will-tell-you-nothing-useful-about-morality https://www.currentaffairs.org/2017/11/the-trolley-problem-w...
  - jsight 9y ago
    I really doubt that the survey was about actually training the cars. It probably has more to do with positioning the marketing messaging around them. AFAIK, the trolley problem isn't something that the engineers consider to actually be a major problem.
    - 2474 9y ago
      Thanks for the insight jsight. I sure hope engineers wouldn't use it. To be fair, it has been a few years since I took the survey so I can't recall if it mentioned what the data was used for. I don't know enough about machine-learning and self-driving technology but my one question would be, where do engineers get that kind of data to feed into a system? At some point a scenario like the Trolley Problem is going to happen (near or distant future) and what kind of data is going to be making that decision to swerve or remain on course?

by wjnc 9y ago

reificator 9y ago

You're right. Developing video games has taught me that a second is a long time. But we still don't even know that Uber didn't detect them that early. The CV system or other sensors could have picked out the pedestrian but something else went wrong instead. I tend to think that if Uber did detect the pedestrian, they would have said something by now. But perhaps they prefer the optics of not having seen her to the optics of seeing her but not braking at all. We simply don't know. Imagine that press conference. > Uber: "After examining the black box on the car we have determined that the car detected the pedestrian but failed to brake." > Reporter: "Was this because the pedestrian was a woman?" > Reporter: "Did the car do this to protect the life of the Uber employee inside?" > Reporter: "Was her credit score too low?"

2474 9y ago
This reminds me of a survey I had taken. I can't remember the survey publisher but it was related to self-driving vehicles. You were presented two pictures (e.g. a picture of school kids crossing a crosswalk and the other a picture of an adult male and female walking their dog across a crosswalk). Presumably you had to choose; would you swerve to hit the adults and dog or would you stay course and hit the kids. I know there is no moral compass for a computer system to determine what it should do in the scenario but I laughed at the quiz because I couldn't believe this is the kind of data that will feed machine-learning algorithms. It just feels bad.
- lowbloodsugar 9y ago
  Getting data on an ethical dilemma. [1] [1] https://en.wikipedia.org/wiki/Trolley_problem https://en.wikipedia.org/wiki/Trolley_problem
  - 2474 9y ago
    Thanks for mentioning what that dilemma was called. For those interested here is an interesting article about the Trolley Problem. https://www.currentaffairs.org/2017/11/the-trolley-problem-will-tell-you-nothing-useful-about-morality https://www.currentaffairs.org/2017/11/the-trolley-problem-w...
- jsight 9y ago
  I really doubt that the survey was about actually training the cars. It probably has more to do with positioning the marketing messaging around them. AFAIK, the trolley problem isn't something that the engineers consider to actually be a major problem.
  - 2474 9y ago
    Thanks for the insight jsight. I sure hope engineers wouldn't use it. To be fair, it has been a few years since I took the survey so I can't recall if it mentioned what the data was used for. I don't know enough about machine-learning and self-driving technology but my one question would be, where do engineers get that kind of data to feed into a system? At some point a scenario like the Trolley Problem is going to happen (near or distant future) and what kind of data is going to be making that decision to swerve or remain on course?