>>/5657/
I agree that a good utility function is (for now) a platonic ideal that can only be uncertainly and somewhat arbitrarily approximated.
I don't think that a system being chaotic has to make it impossible to apply a utility function. It adds variance, but it doesn't necessarily preclude expected values.
And I don't think that such an approximation has to be useless. To be useless, it's not enough if it's imperfect - it has to do worse at satisfying the platonic utility function than other workable systems. I think QALY/$ estimates are good enough to be useful.
You can't prove with 100% certainty that it's good, but that goes for everything.