From: Wei Dai (firstname.lastname@example.org)
Date: Tue Sep 16 2003 - 17:28:30 MDT
On Tue, Sep 16, 2003 at 06:09:06PM -0400, Eliezer S. Yudkowsky wrote:
> I haven't looked into the result you refer to, Wei Dai, but my initial
> impression is that it assumes infinite degrees of freedom in both U(x) and
> P(x). Leaving aside the former, the latter, at least, is usually assumed
> to normatively obey Bayesian probabilities, and cognitively it obeys
> certain loose non-Bayesian rules. We are surprised when we see that under
> certain circumstances people assign subjective likelihoods P(A&B) > P(A),
> but having discovered this, we can then predict that most people will do
> it most of the time under those conditions. So there are not infinite
> degrees of freedom in P(x), either normatively or cognitively, and if you
> use this constraint to construct U(x) you will not find infinite relevant
> degrees of freedom in U(x) either. Of course I am only reasoning
> intuitively here, and I may have gotten the math wrong.
>From what I remember, you can always find an infinite set of pairs of
U' and P' that satisfy the constraints of decision theory, and P' != P.
I think in most cases you will get different volitional orderings from
these U'. I can look up the math details if you're interested.
> But, mostly my reply is that I'm not using the economist assumption that
> we have to construct U(x) and P(x) by looking exclusively at people's
> choices. From a volitionist standpoint, what I would like to do is open
> up people's heads and look at their mind-state, figure out what systems
> are working, what they contain, what actual system is producing the
> choices, and then, having this functional decomposition, ask which parts
> have U(x) nature and which parts have P(x) nature, bearing in mind that
> they will overlap. Existing cognitive psychology actually goes quite a
> ways toward doing this.
So how do you determine whether the subject volitionally wants you to
open up his head and look inside?
But putting that aside, the main point I want to make is that I don't
see what difference exists between the utility function and the
probability function that makes you want to respect a person's utility
function but not his probability function. Both functions need to
satisfy certain constraints according to decision theory, but other
than that they are completely subjective and arbitrary. Why not have
the Friendly AI implement an alternative moral theory where the AI
makes decisions for a person according to his own probability function
but substituting the AI's idea of a "correct" utility function for the
subject's? That makes just as much sense to me as volitionism.
>From the subject's perspective, both of these approaches are equally
bad. In both cases the AI is doing things that hurt his expected
utility. Why should the fact that the AI respects his utility function
function be more of a consolation than the fact that the AI respects
his probability function?
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT