A permutation argument for comparing utility functions

post by Stuart_Armstrong · 2017-04-27T15:01:00.000Z · LW · GW · 2 comments

Contents

    Permutation
  Another type of uncertainty
    Consequences
None
2 comments

When doing intertheoretic utility comparisons, there is one clear and easy case: when everything is exactly symmetric.

This happens when, for instance, there exists and such that and there exists a map , such that is the identity (hence is an involution) and for all ), .

Note that this implies that , so is essentially a 'reflection'.

Then, since everything is so symmetric, we can say there is no way of distinguishing from , so the correct approach is to maximise .

See the following graph, with the strategy to be followed marked in red:

Permutation

Symmetry is good as far as it goes, but is very fragile. It doesn't say anything about what happens when and , for instance.

There is an argument, however, that resolves the unequal probability cases and extends the results to non-symmetric cases.

Consider for instance the case where and are as follows, for strategies in :

Nothing obvious springs to mind as to what the best normalisation process is -- the setup is clearly unsymmetrical, and four of the five options are on the Pareto boundary. But let's rescale and translate the utilities:

This is still unsymmetrical, but note that and have the same values: , , , , and .

Thus there is a permutation such that for all , .

Another type of uncertainty

This permutation allows us to transform the uncertainty about one's own values (which we don't know how to handle) to other types of uncertainty (which we can).

How so? Let be another copy of , but with different labels, and let be the identity map that re-assigns each element to its original label.

Then instead of seeing and as different utility functions on , we can see them as both being the same utility function on , with uncertainty over a map from to . This map is with probability and with probability .

Thus the agent should see itself as a -maximiser, with standard uncertainty over the map (this could be seen as uncertainty over the consequences of choosing a strategy). This means it will maximise , as long as and are related by a permutation on .

In this particular case, if and are close to , this means that the red strategy will be selected:

Note that permutations includes the symmetric case where is an involution, so the symmetric argument now extends to cases where .

Consequences

For permutations, this argument claims that the correct approach is to use Individual normalisations. Indeed, every normalisation presented here would reach the same result.

This doesn't prove that individual normalisation is necessarily correct -- you can imagine a general system that only give individual normalisations on permutation problems -- but is suggestive none the less.

2 comments

Comments sorted by top scores.

comment by owencb · 2017-04-27T15:12:50.000Z · LW(p) · GW(p)

I'm not sure I've fully followed, but I'm suspicious that you seem to be getting something for nothing in your shift from a type of uncertainty that we don't know how to handle to a type we do.

It seems to me like you must be making an implicit assumption somewhere. My guess is that this is where you used to pair with . If you'd instead chosen as the matching then you'd have uncertainty between whether should be or . My guess is that generically this gives different recommendations from your approach.

Replies from: Stuart_Armstrong
↑ comment by Stuart_Armstrong · 2017-04-27T15:58:32.000Z · LW(p) · GW(p)

Nope! That gives the same recommendation (as does the same thing if you pre-compose with any other permutation of ). I thought about putting that fact in, but it took up space.

The recommendation given in both cases is just to normalise each utility function individually, using any of the methods that we know (which will always produce equivalent utility classes in this situation).