I’m really curious to know what you mean by ‘terminal meta-values’. Would you mind expanding a bit, or pointing me in the direction of a post which deals with these things?
No, I’m perfectly OK with adjusting terminal values in certain circumstances. For example, turning a Paperclipper into an FAI is obviously a good thing.
EDIT: Of course, turning an FAI into a Paperclipper is obviously a bad thing, because instead of having another agent working towards the greater good, we have an agent working towards paperclips, which is likely to get in the way at some point. Also, it’s likely to feel sad when we have to stop it turning people into paperclips, which is a shame.
I’m really curious to know what you mean by ‘terminal meta-values’. Would you mind expanding a bit, or pointing me in the direction of a post which deals with these things?
Say, whether it is ever acceptable to adjust someone’s terminal values.
No, I’m perfectly OK with adjusting terminal values in certain circumstances. For example, turning a Paperclipper into an FAI is obviously a good thing.
EDIT: Of course, turning an FAI into a Paperclipper is obviously a bad thing, because instead of having another agent working towards the greater good, we have an agent working towards paperclips, which is likely to get in the way at some point. Also, it’s likely to feel sad when we have to stop it turning people into paperclips, which is a shame.