May be we could use it in more recursive form. I may ask AI X what would do AI X if it have an opportunity to do bad think Y?
Would you kill me if you have a gun? While it could lie here, we may have other ways to test in general honesty.
May be by using to different AIs: I ask AI X what would do AY Z in case if it has access to Y.
Or by running many trails.
May be we could use it in more recursive form. I may ask AI X what would do AI X if it have an opportunity to do bad think Y? Would you kill me if you have a gun? While it could lie here, we may have other ways to test in general honesty. May be by using to different AIs: I ask AI X what would do AY Z in case if it has access to Y. Or by running many trails.