Yeah, the important part of it would be designing selection criteria so that only a useful universe is presented for inspection. (Of course an infinite number of universes will meet any set of criteria, but that's fine - since they're all useful then just take the first.)

So, one criterion would be that a question is injected into any universe which contains things which appear to have intelligent characteristics, and the response would have to be something which could be parsed sensibly in this universe.

I think that chaos theory can be handled if you have infinite resources and a precise use case in mind.