Stupid question: what about defining "lying" as "saying something that, if accepted, moves the believer away from the liar's worldmodel"? (Which I think matches general intuition.)
Then you do get information once you know someone always lies (and has a correct worldmodel) because you know to update the opposite direction. I don't know the impact on these puzzles.
what about defining "lying" as "saying something that, if accepted, moves the believer away from the liar's worldmodel"?
Then the definition becomes kinda-contradictory: The entity would sometimes tell the truth, because that would be the best choice to divert and ruin your model.
So "always lies in a simple way" would become more like "always unhelpful in a superintelligently evil way."
That wouldn’t make it contradictory, it would make “always lying” just as effective a strategy as in the original problem statements. It’s a general problem that “being known to always lie” makes (the goals of) deception harder, and my definition just preserves that problem.
Comments
Stupid question: what about defining "lying" as "saying something that, if accepted, moves the believer away from the liar's worldmodel"? (Which I think matches general intuition.)
Then you do get information once you know someone always lies (and has a correct worldmodel) because you know to update the opposite direction. I don't know the impact on these puzzles.
Then the definition becomes kinda-contradictory: The entity would sometimes tell the truth, because that would be the best choice to divert and ruin your model.
So "always lies in a simple way" would become more like "always unhelpful in a superintelligently evil way."
That wouldn’t make it contradictory, it would make “always lying” just as effective a strategy as in the original problem statements. It’s a general problem that “being known to always lie” makes (the goals of) deception harder, and my definition just preserves that problem.
In some contexts in computer science and computer security, this is also called "answering adversarially".