AIs, Honesty, Murder and Trolleys

[vicky_molokh]

There's definitely a delusion involved. The question was whether the delusion was accompanied by the removal of safety systems of law-abiding programming that would be required to actually commit a mass murder (well, try to).

[jeff_wilson]

If the Trolley Dillemma is really applicable, the thing to do would be to research plane crashes in France where the pilot was able to divert the plane from killing more people toward less, and see if his estate or employer was held liable for the deaths on the ground but not on the plane.

[Anthony]

Originally Posted by jeff_wilson

If the Trolley Dillemma is really applicable, the thing to do would be to research plane crashes in France where the pilot was able to divert the plane from killing more people toward less, and see if his estate or employer was held liable for the deaths on the ground but not on the plane.

Not really equivalent; that would just be 'liability based on who you hit', because he has responsibility for the plane. The trolley dilemma involves someone who has no prior responsibility for the train.

[jeff_wilson]

Originally Posted by Anthony

Not really equivalent; that would just be 'liability based on who you hit', because he has responsibility for the plane. The trolley dilemma involves someone who has no prior responsibility for the train.

If the person is standing in the train yard at the switch controls, it's presumably because they *do* have prior responsibility for the train. But it's not particularly important that they be unconnected with the train, just that they not be responsible for the situation to begin with, and finding the people previously bound and laid in the way takes care of that. Similarly, if the pilot is not held responsible for the people on board dying, that indicates he's not responsible for the crashing condition, only for the choice of where to crash.

[Keiko]

Originally Posted by vicky_molokh

There's definitely a delusion involved. The question was whether the delusion was accompanied by the removal of safety systems of law-abiding programming that would be required to actually commit a mass murder (well, try to).

Hm, so that programming couldn't be suppressed or corrupted even temporarily (like temporary insanity for a human) maybe living the AI fundamentally broken mentally afterward (an insanity/wracked with guilt like state).