Re: [UAI] Fuzzy sets vs. Bayesian Network

[Stuart Russell]

Kathy,
Good answer. I believe something along these lines can be made to work.
Of course, you *don't* get the fuzzy rules, which are truth-functional
and hence insensitive to the nature of dependence between propositions.
(See Elkan etc., cited in AIMA.) But an interesting calculus might
emerge. Notice that making the proposition be "X says P in context C"
brings in a distribution over contexts and speakers as well as the value
of P itself. I drafted some of this for inclusion in AIMA but
decided against it, not wishing to seem too probabilist-imperialist
with Lotfi in the next-door office.
Stuart