I bet a median-utility maximizer can be exploited. But I don’t believe one can be exploited by a Pascal’s mugging. What makes a Pascal’s mugging a Pascal’s mugging is that it involves a very low probability of a very large change in utility.
I’m not sure this is a useful question. I mean, if you choose the (1-p) quantile (I’m assuming this means something like “truncate the distribution at the p and 1-p quantiles and then take the mean of what’s left”, which seems like the least-crazy way to do it) then any given Pascal’s Mugging becomes possible once p gets small enough. But what I have in mind when I hear “Pascal’s Mugging” is something so outrageously improbable that the usual way of dealing with it is to say “eh, not going to happen” and move on (accompanied by a delta-U so outrageously large as to allegedly outweigh that), and I take Houshalter to be suggesting truncating at a not-outrageously-small p, and the two don’t really seem to overlap.
I bet a median-utility maximizer can be exploited. But I don’t believe one can be exploited by a Pascal’s mugging. What makes a Pascal’s mugging a Pascal’s mugging is that it involves a very low probability of a very large change in utility.
Do you believe that the 99.999-percentile by utility-ordered outcome count can be Pascal-mugged? How about 90%? Where is the cut-off?
I’m not sure this is a useful question. I mean, if you choose the (1-p) quantile (I’m assuming this means something like “truncate the distribution at the p and 1-p quantiles and then take the mean of what’s left”, which seems like the least-crazy way to do it) then any given Pascal’s Mugging becomes possible once p gets small enough. But what I have in mind when I hear “Pascal’s Mugging” is something so outrageously improbable that the usual way of dealing with it is to say “eh, not going to happen” and move on (accompanied by a delta-U so outrageously large as to allegedly outweigh that), and I take Houshalter to be suggesting truncating at a not-outrageously-small p, and the two don’t really seem to overlap.