This is a linkpost for

Franz Dietrich & Brian Jabarian. Expected value under normative uncertainty. Preprint.

Maximising expected value is the classic doctrine in choice theory under empirical uncertainty, and a prominent proposal in the emerging philosophical literature on normative uncertainty, i.e., uncertainty about the standard of evaluation. But how should Expectationalism be stated in general, when we can face both uncertainties simultaneously, as is common in life? Surprisingly, different possibilities arise, ranging from Ex-Ante to Ex-Post Expectationalism, with several hybrid versions. The difference lies in the perspective from which expectations are taken, or equivalently the amount of uncertainty packed into the prospect evaluated. Expectationalism thus faces the classic dilemma between ex-ante and ex-post approaches, familiar elsewhere in ethics and aggregation theory under uncertainty. We analyse the spectrum of expectational theories, showing that they reach diverging evaluations, use different modes of reasoning, take different attitudes to normative risk as well as empirical risk, but converge under an interesting (necessary and sufficient) condition.

There's a couple of interesting things in this one. One is the taxonomy of approaches to normative uncertainty, which lays out dimensions/variables along which a subclass of methods for resolving normative uncertainty can differ. The other is that teased condition under which these methods converge.

I won't keep you waiting or make you read the whole thing. The theorem is that they converge when the expected utility hypothesis is true, i.e. in the case that vNM utility theory accurately models rationality.

Be advised that this is an unpublished preprint, and I haven't read it carefully, so I can't vouch for its correctness. Shared because it looked interesting and is trying to make progress towards resolving disagreements over how to resolve normative uncertainty.

Full-text here.




Sorted by Click to highlight new comments since:

I think a major limitation is that they assume (real-valued) comparability between theories/their valuations. You'd need to motivate particular comparisons between theories, and I don't know that this can always be done.

Curated and popular this week
Relevant opportunities