By misaligned I mean an AGI that isn't aligned with the ACTUAL values of most humans.
And by taking over the world I mean the news would say so.
If you have recomendations that would make this description more useful I'm open to hear them.
Updated twice 08/02/2023: The definition of misaligned was to not being aligned with a single human and then was to not being aligned with any set of humans.
Feb 8, 12:42am: Will a misaligned AI take over the world? → Will a misaligned AGI take over the world?
Let’s say it proves to be too vulnerable to an EMP, and therefore requires humans to stand around as a “backup”. Would that count as a “No”?
@ThomasKwa What if the AI convinces the AI's creator that that is what they want even if the AI was initially "misaligned"?
fuck I ctrl+enter instead of shift+enter
for the first point: you are right, I want to consider that alignment so I changed the description. I also deleted the "intent alignment" part because I don't actually know the consensus of the term.
for the second point: if they regret it I wouldn't consider it properly aligned.
@patodesu Doesn't it make sense that the AI would modify the human's brain? Everyone regrets things sometimes so what is the criteria? Eg. they write an article saying they regret it or what?
@ZZZZZZ First question: Not necessarily, but I will change the description so that will not be a problem anymore.
Second question: I consider that to actually align an AI means to give it objectives that you will not regret giving it. The thing is, I don't know how the criteria would be to know that and now that I think about it, what I really care is the chances of align it with "human values", AKA the most ambitous definition of alignment.
So sorry but I'll change the description again.
@tailcalled well that wouldn't be misaligned i think. Just in case in the description I also put AGI