Re: my current position (was: AUDA)

From: Wei Dai <weidai.domain.name.hidden>
Date: Fri, 11 Jan 2002 20:16:13 -0800

On Fri, Jan 11, 2002 at 04:59:47PM -0800, hal.domain.name.hidden wrote:
> I'm having a lot of trouble understanding this view.

Thanks for taking the time to write the questions. I hope this response
helps.

> Why should you care more or less about slow to compute universes?

I don't see any reason to care more or less about slow to compute
universes, so I care about them as much as fast to compute universes. But
in general it's an arbitrary subjective choice.

> What kinds of considerations would influence your decision to care about
> such universes?

For example, you might be living inside a simulation and also in a "real"
world (for whatever definition of real), and you can decide that you don't
care about what happens in simulations as much as in "real" universes.
But you don't need a rational justification for it, in the same way that
you don't need a rational justification for preferring, say, abstract art
to representational art.

> Isn't it an empirical question which prior obtains (speed vs universal)?

No, I argue it's not, it's a matter of preference. Note that Jurgen
Schmidhuber also argues that it's not, but I think his position is that
it's a matter of logical necessity. (Have you read his TOESV2 paper yet,
BTW?) Note that in both of our approaches you'd do the same things if you
adopt the same prior. But our justifications/interpretations are
different.

> You want to maximize your gains, so you try to figure out from reason
> and observation which prior is true. For example you could build a small
> quantum computer and see if worked. If not that would suggest that the
> speed prior is true, if it does work that suggests the universal prior
> is true.

Let's see how this would work in my approach. If you adopt the speed
prior, then you care about the fast-to-compute universes so little that
you wouldn't even bother building a small quantum computer as an
experiment. Even if you observed someone else's quantum computer working,
you'd continue to act as if quantum computing was impossible because you
care more about the universes where that experience was the result of a
hallucination.

In Schmidhuber's approach, you'd think that the probability you being in a
universe where the quantum computer would work is so small that it's not
worth trying. If you observed someone else's quantum computer working,
you'd think that it happened because you hallucinated.

(BTW, Jurgen, if I'm misinterpreting please let me know.)

Now suppose you adopt the universal prior instead, and assume that the
measure of the fast-to-compute universes is now small but significant
compared to the slow-to-compute universes. In my approach, you'd do an
quantum computing experiment in order to choose different actions in
different universes. (I.e. so you don't have to make the same decision in
both kinds of universes, but can condition your decision based on which
kind of universe you're in.) If you observe the quantum computer not
working, you'd think that you can no longer affect the slow-to-compute
universes, and therefore do not have to consider them any more in making
your decisions. (But you know that versions of you in other universes have
observed the quantum computer working, and that they can stop considering
the fast-to-compute universes.)

> Suppose you observe that quantum computers don't work. What does that
> mean in your formulation? Does it mean that you have decided to care
> about a certain kind of universe? Why should this fact change what you
> care about?

No it doesn't change what you care about, it only changes what universes
you can affect, and by extension which ones you can no longer
affect and don't have to worry about anymore. So the point of making
observations (and remembering them) is to minimize the measure of
universes you will affect and have to consider when you choose a course of
action.
Received on Fri Jan 11 2002 - 20:19:00 PST