Sorceress

The Sorceress Schema

Aug 15th, 2016
225
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
  1. The Sorceress Schema
  2. --------------------
  3.  
  4. My proposed solution to the thought experiments posed at http://moralmachine.mit.edu/
  5.  
  6. Basically, the situation is a self-driving vehicle with failed brakes. The vehicle must decide whether to go straight on or swerve, and people will lose their lives no matter which decision is made. The machine is faced with a moral dilemma of which action to take, and which lives to save.
  7.  
  8.  
  9. FIRST THOUGHTS
  10. --------------
  11. Firstly, I think it's important that AI is predictable. In real-world scenarios there will always be time for pedestrians to react and possibly save themselves, and predictability will help people make the best split decision they can.
  12.  
  13. If the vehicle can make a decision leading to no deaths, then that is the best option. For example, an empty car crashing itself into a wall is better than it crashing into pedestrians. An occupied car moving into a clear lane is better than it crashing into pedestrians.
  14.  
  15.  
  16. EMPTY CAR
  17. ---------
  18. If there is no wall to crash an empty car, and there are pedestrians on both sides of the road, then a decision has to be made which people to run into.
  19.  
  20. Pedestrians crossing red lights have the lowest moral considerability, imo. They put themselves into a life-risking situation. So if a vehicle has to choose whose lives to put at most risk, then I think it should be red-light crossers. Quantity, age, and species should not factor. Green-light crossers should not be sacrified because others have chosen to put themselves at risk.
  21.  
  22. So pedestrians with Red-light + pedestrians with Green-light -> car should take the red-light lane.
  23.  
  24. If the two lanes are equal colour, then both sets of pedestrians are of equal moral considerability. So in this situation I think the vehicle should aim to save the most lives. And if numbers are equal, the vehicle should carry straight on. (Predictability and maybe the people on that side of the road should have looked more carefully before crossing.)
  25.  
  26.  
  27. OCCUPIED CAR
  28. ------------
  29. If the vehicle has passengers then there are more lives to consider. I would place vehicle passengers below green-light pedestrians in terms of moral considerability, but higher than red-light pedestrians. Passengers have ultimately chosen to ride in this vehicle which has the potential to kill, They are responsible for the car being where it is. So I think this is a reasonable price to pay.
  30.  
  31. So if there is an empty lane, then move to that lane.
  32. If both lanes have pedestrians, then move to the red-light lane
  33. - if both lanes are the same colour, then aim to save the most lives.
  34. - - if the numbers are equal, then go straight on.
  35.  
  36. If there is a wall + green-light pedestrians, then crash into the wall.
  37. If there is a wall + red-light pedestrians, then move to the red-light lane.
  38.  
  39. SUMMARY
  40. -------
  41. To put this into algorithmic form:
  42.  
  43. if(there is a solution with no deaths) {
  44. use solution with no deaths;
  45. }
  46. else {
  47. // value green-light pedestrians > vehicle passengers > red-light-pedestrians
  48. if (choice is between different "value" groups) {
  49. car kills the lowest value group;
  50. }
  51. else {
  52. if (choice is between different sized groups) {
  53. car kills smaller group;
  54. }
  55. else {
  56. car goes straight on;
  57. }
  58. }
  59. }
  60.  
  61. -----
  62. (c) Sorceress, 2016
RAW Paste Data