0 comments
Comments sorted by top scores.
comment by MondSemmel · 2022-04-03T14:04:10.990Z · LW(p) · GW(p)
If your math is telling you this doesn't make sense, remember that math is just a map
Or alternatively, your conclusion might be (not even) wrong.
I agree that there's far more cooperation and far less defection than one would expect if all of human interaction were just true one-shot prisoner's dilemmas; but your explanation for why doesn't ring true to me.
In fact, mentioning "friendly agent C" is kind of begging the question, isn't it? It's presupposing that there already exist powerful cooperators.
Replies from: eg↑ comment by eg · 2022-04-03T14:38:24.194Z · LW(p) · GW(p)
Maybe true one-shot prisoner's dilemmas aren't really a thing, because of the chance of encountering powerful friendliness.
We have, for practical purposes, an existence proof of powerful friendliness in humans.
↑ comment by Aiyen · 2022-04-17T20:40:01.130Z · LW(p) · GW(p)
Privileging the hypothesis. There could be powerful friendly agents who punish unfriendliness, but unless we figure out how likely they are, the mere possibility isn't meaningful. There could also be powerful agents that will punish us unless we harm the weak, but merely knowing that this isn't impossible isn't a good reason to do so.