As with many anthropic considerations, there is a serious problem determining the reference class here. Generally an appropriate reference class is “somebody sufficiently like you”, and then compute weightings for some parameter that varies between universes and affects the number and/or probability of observers.
The trouble is that “sufficiently like you” is a uselessly vague specification. The most salient reference class seems to be “people considering removing a fence very much like this one”. But that’s no help at all! People in other universes who already removed their universe’s fence are excluded regardless of whether they lived or died.
Okay, what about “people who have sufficiently close similarity to my physical and mental make-up at (time now)”? That’s not much help either: almost all of them probably have nothing to do with the fence. Whether or not the fence is deadly will have negligible effect on the counts.
Maybe consider “people with my physical and mental make-up who considered removing this fence between (now minus one day) and (now), and are still alive”. At this point I consider that I am probably stretching a question to get a result I want. What’s more, it still doesn’t help much. Even comparing universes with p=0 of death to p=1, there’s at most a factor of 2 difference in counts for the median observer. Given such a loaded question, that’s a pretty weak update from an incredibly tiny prior.
As with many anthropic considerations, there is a serious problem determining the reference class here. Generally an appropriate reference class is “somebody sufficiently like you”, and then compute weightings for some parameter that varies between universes and affects the number and/or probability of observers.
The trouble is that “sufficiently like you” is a uselessly vague specification. The most salient reference class seems to be “people considering removing a fence very much like this one”. But that’s no help at all! People in other universes who already removed their universe’s fence are excluded regardless of whether they lived or died.
Okay, what about “people who have sufficiently close similarity to my physical and mental make-up at (time now)”? That’s not much help either: almost all of them probably have nothing to do with the fence. Whether or not the fence is deadly will have negligible effect on the counts.
Maybe consider “people with my physical and mental make-up who considered removing this fence between (now minus one day) and (now), and are still alive”. At this point I consider that I am probably stretching a question to get a result I want. What’s more, it still doesn’t help much. Even comparing universes with p=0 of death to p=1, there’s at most a factor of 2 difference in counts for the median observer. Given such a loaded question, that’s a pretty weak update from an incredibly tiny prior.