I don’t feel like I understand the intense academic focus on choice modeling and similar over asking people directly for the relative values, particularly with something like probability distributions.
I get that the latter requires more sophisticated users, but it also provides far more precision. For a lot of decision making, you really want some simple real utility function.
In EA I think many of our main decision makers are sophisticated enough for the latter methods. Also, for those not yet, I’d be curious to try methods like giving some representative samples the necessary training/education, then doing elicitation on these enlightened groups.
If there really isn’t much literature or tooling on numeric/distribution/utility-function elicitation, it seems like it should be really low hanging fruit.
I found this really useful, kudos for writing it.
I don’t feel like I understand the intense academic focus on choice modeling and similar over asking people directly for the relative values, particularly with something like probability distributions.
I get that the latter requires more sophisticated users, but it also provides far more precision. For a lot of decision making, you really want some simple real utility function.
In EA I think many of our main decision makers are sophisticated enough for the latter methods. Also, for those not yet, I’d be curious to try methods like giving some representative samples the necessary training/education, then doing elicitation on these enlightened groups.
If there really isn’t much literature or tooling on numeric/distribution/utility-function elicitation, it seems like it should be really low hanging fruit.