From: Wei Dai (weidai@weidai.com)
Date: Fri Feb 14 2003 - 20:07:38 MST
On Thu, Feb 13, 2003 at 01:04:34AM -0500, Eliezer S. Yudkowsky wrote:
> You have a very nice definition of decisionmaking as a policy which
> estimates the desirability of the multiverse given that the outcome of a
> decision process is X, then chooses the option which results in greatest
> moral value.
>
> Given this perfectly good definition, which suffices for all conceivable
> activities of an intelligence, why do you feel the need to talk about this
> additional entity called "probabilities"?
During what you call "estimation of desirability" you need to handle some
uncertainties. You're not sure what the current state of the multiverse
is. You're not sure which observer-moments in the multiverse you might be.
You're not sure what are the consequences to the multiverse of the
candidate course of action. I call the numbers that represent the degrees
of these uncertainties "probabilties" since they are numbers between 0 and
1 and are used as weights for computing expected utility. Does that answer
your question?
This archive was generated by hypermail 2.1.5 : Fri Feb 14 2003 - 20:10:04 MST