3 ms·

Here is a link where the Director of the CIA openly and proudly admits to killing people based on meta-data: https://www.youtube.com/watch?v=UdQiz0Vavmc https:/

by Phlarp 11y ago

Here is a link where the Director of the CIA openly and proudly admits to killing people based on meta-data: https://www.youtube.com/watch?v=UdQiz0Vavmc https://www.youtube.com/watch?v=UdQiz0Vavmc Surely you don't think they have gymnasiums full of junior analysts trying to manually correlate phone call metadata? At some level an algorithm decides someone specific crosses the probability threshold for our definition of "terrorist" and it's all rubber stamps from there until the hellfire blows up a wedding[1], funeral[2] or hospital[3]. [1]: http://www.democracynow.org/2014/2/21/turning_a_wedding_into_a_funeral http://www.democracynow.org/2014/2/21/turning_a_wedding_into... [2]: http://www.theatlantic.com/international/archive/2013/10/drone-attacks-at-funerals-of-people-killed-in-drone-strikes/280821/ http://www.theatlantic.com/international/archive/2013/10/dro... [3]: http://www.nytimes.com/2015/10/05/world/asia/doctors-without-borders-says-it-is-leaving-kunduz-after-strike-on-hospital.html http://www.nytimes.com/2015/10/05/world/asia/doctors-without...

xixi77 11y ago
Well, IMO there is a big difference between humans looking at algorithm-provided evidence (even if it's often rubber-stamping), and drones acting on that evidence directly.
grkvlt 11y ago
Yes, but your examples [1,2,3] are neither the desired, nor the likely outcome. In most cases, the correct (not nescessarily the best or morally right) target is eliminated. In fact, I suspect (and would be interested in data that could prove or disprove this) that the erroneous outcome types [1,2,3] you list are _less_ likely with drones and automated systems than with humans on the front line and the attendant stress and confusion implicit in that situation...
- Phlarp 11y ago
  In [1] and [2] the "target" of the strike was eliminated and the dozen or so civilians that died in each are considered acceptable collateral damage. I suspect this is also the case for [3], however they are unlikely to own up to that publicly

by Phlarp 11y ago

xixi77 11y ago

Well, IMO there is a big difference between humans looking at algorithm-provided evidence (even if it's often rubber-stamping), and drones acting on that evidence directly.

grkvlt 11y ago

Yes, but your examples [1,2,3] are neither the desired, nor the likely outcome. In most cases, the correct (not nescessarily the best or morally right) target is eliminated. In fact, I suspect (and would be interested in data that could prove or disprove this) that the erroneous outcome types [1,2,3] you list are _less_ likely with drones and automated systems than with humans on the front line and the attendant stress and confusion implicit in that situation...

Phlarp 11y ago
In [1] and [2] the "target" of the strike was eliminated and the dozen or so civilians that died in each are considered acceptable collateral damage. I suspect this is also the case for [3], however they are unlikely to own up to that publicly