How about, "What do I think /I/ want out of society, and how do I optimize for that?"?
In theory that might be the best way of going about things, but if it doesn't generalize well to other people, you're unlikely to get others on board with it, which limits the usefulness of framing the question that way.
But surely DataPacRat's is the correct question. (Of course, if what DataPacRat really desires is that other people get what they want, then it's hard to distinguish the questions.) Once answered, and in the optimisation phase, then we consider how best to frame discussions of the issues to convince other people that they want it too (or whatever is most effective).
Politics is the mind-killer; but rationality is the science of /winning/, even when dealing with political issues.
I've been trying to apply LessWrong and Bayesian methods to the premises and favored issues of a particular political group. (Their most basic premise is roughly equivalent to declaring that Iterated Prisoner's Dilemma programs should be 'nice'.) But, given how quickly my previous thread trying to explore this issue was downvoted into disappearing, and many of the comments I've received on similar threads, I may have a rather large blind spot preventing me from being able /to/ properly apply LW methods in this area.
So I'll try a different approach - instead of giving it a go myself again, I'll simply ask, what do /you/ think a good LW post about liberty, freedom, and fundamental human rights would look like?