Posts

Sorted by New

Wiki Contributions

Comments

Sorted by
sitomin7243610

Corollaries:

Honesty

  • If you want to become more honest and less self-deceiving, acquire power
  • If you want to make other people more honest and less self-deceiving, provide them with power (including power to protect themselves from you)
  • If you know someone who is more powerful than you but cant guarantee an upper bound on their power (and future power), then occlumency no longer works

Unboundedness

  • If you want an unlimited amount of power (such as a utility maximiser), there will almost always be coalitions of people more powerful than you against whom self-deception works
  • As long as there exist (hostile) coalitions of people unboundedly more powerful than you, completely removing self-deception from yourself is impossible

More than just yourself

  • If you want more examples of honesty and lack of self-deception available to you, ask powerful people to speak about their life experience. If you want these examples to be public, make them public
  • If you want two agents hostile to each other to both simultaneously be honest and not self-deceiving, provide them defensive rather than offensive power
  • If you want to achieve world peace, consider building defensive but not offensive power for every level of self-organisation - individual, family, ideological group, geographic group, etc etc

Time

  • If you don’t want someone to learn skills of dishonesty and self-deception, provide them with power as early as possible
  • If you don’t want to learn skills of dishonesty and self-deception, acquire power as early as possible

Hiding

  • If you want to acquire power without dishonesty or self-deception, ensure that your mental state and all causally downstream changes to world state are indistinguishable from noise in the eyes of more powerful actors.

I’m sure you can find real world examples of most of these.

Also there might exist edge cases where some of the above corollaries don’t hold, those edge cases are worth exploring.

Also: this post is refreshing because it is not about AI and has the same vibes as Lesswrong 2008 (back when LW was actually good). I made an alt just to reply to it.