Specifically, here is a new paper on Friendly AI I'd like to understand first.
UPDATE: A nice stimulating paper [it's interesting, how they use Wiki too, in a bit unusual fashion]. The main problem of Yudkowsky's approach here is that it is too humanity-centric. Basically, if we have parents (us) and their kids (AI), certainly things cannot be entirely determined by the parents' interests, especially in a situation where kids are so much smarter.
Although, for creators with a more "Oriental" system of values this might seem appropriate...
