Re: Bruno's UDA argument

From: Bruno Marchal <>
Date: Sat, 27 Jul 2002 18:19:46 +0200

I comment the first part of Hal's last UDA post.
(I put BM and HF because the archive does not print correctly my
quotes apparently)

At 10:25 -0700 22/07/2002, Hal Finney wrote:

>Here is my response to Bruno's UDA argument as described in the links
>I agree with the general idea, that if we assume that a UDA program
>exists in some sense (Platonic or otherwise), then each observer moment
>is instantiated multiple times (probably a countably infinite number of
>times) throughout the structure generated by the UDA. In that case we
>have to accept that our mental experiences are not unique to some universe
>(by "universe" I mean the output of one program being simulated by the
>UDA) but are present in infinitely many of them.


>However there are a few places where I would expand upon the steps, and
>take issue with some of them. First, as far as the basic idea of COMP,
>I do agree that functional substitution is enough for consciousness
>indistinguishable from my own. I would expand on this to explain that
>the reason for this belief is not due to philosophy, but to biology.

To biology + some philosophy I would say. That's ok, my first feeling that
comp is plausible comes also from biology. Even microbiology ...

>The way the brain works appears to be a matter of processing information
>via electrochemical signals transmitted by neural cells. Each such cell
>is acting as an information processor. The collection of these cells
>in the brain takes in information on incoming nerves, and sends out
>commands on outgoing nerves. Everything is information. So it seems to
>me that replacing this system with something that can process information
>in a functionally equivalent way would produce an equivalent system.
>If I didn't know that the brain worked in this way, I would probably
>not accept COMP.

A belief in a reductionist account of cells, or even molecules, or even
just space-time-energy would have been enough for a reductionist
microbiologist, or chemist or physicist. But ok, knowledge
of the brain gives the feeling our level of substitution can be relatively
high. That help conveying the idea. But in the long run it is preferable
not having any *a priori* on the level. Comp assumes only the existence of
the level. (later we will have that we can only bet on a level).

>However, based on some arguments which Wei Dai gave (and which I'm not
>sure if he accepts), I believe there are some aspects of consciousness
>which are not fully captured by a purely functional substitution.
>Specifically, as we get into the questions about duplications, I
>believe that some details of the functional substitution may affect the
>first-person likelihood of experiencing that implementation.

If the details are functional and "digitalizable", then this would be
equivalent to saying that the level of substitution has not been chosen
correctly. If those details are not "digitalizable", this is equivalent
of rejecting comp. I think.

>Issues such
>as the speed, size and degree of redundancy of an implementation may
>affect how much "measure" it has, that is, how much of a contribution
>it makes to the space of all my first-person experiences.

I would agree, if the word "implementation" was not ambiguous in our
context. I agree if by implementation, you mean any digitalizable
implementation. Those belongs to UD*. (Trace of the UD).

>If we jump to the end of Bruno's argument, step 10 or 11, and accept
>that we are instantiated in multiple forms, it is obvious that not
>all ways we could be instantiated are equally probable. Otherwise the
>universe we observe would be chaotic. So somehow it must be the case
>that implementations which exist in simple and lawful universes are more
>probable (in the first person sense) than those in random universes.

I agree. Modulo the change of "equally probable" (at the end of the first
sentence) by "relatively equally probable".

>Given that we accept the reality of different measures for different
>implementation states, I think we need to consider the possibility that
>not all functional substitutions have equal measure (i.e. first-person
>probability). I won't go into the details here, but I think a detailed
>argument shows that different sizes and speeds of the functional
>substitution must make a difference in measure.

"relative speed". The invariance lemma makes the 3-person DU speed
entirely irrelevant. But relative speed is measure in internal space
time, the one which we must shown (by UDA) to emerge from the many
relative situation/history generated by the DU.

>Several of Bruno's questions raise the idea of quantifying the first
>person probability of various duplication scenarios. In the comment
>preceding question 7 in,
>he writes:
>> BM: The way to quantify the indeterminacy is the unknown. Although we can
>> argue that the {W, M} duplication gives a sort of perfect 1-coin, simple
>> probabilty reasoning leads quickly to hard problems.
>HF:I think Bruno agrees here that you can't just say that if you have 2
>duplicates made, you therefore have a 50% first-person probability of
>experiencing each of them. We have discussed various paradoxes and
>problems that arise with this kind of reasoning.

Actually I tend to believe that perfect duplication in distinguishable
environment gives a 1/2 uniform probability. But I don't use this in the

>Nevertheless he asks in questions 7 and 8, *if* we accept a 50%
>probability in one experiment, will we also see it in another? I am not
>sure how to answer these, because I don't have a good understanding of
>how to begin to quantify the probabilities. So I cannot answer in the
>affirmative, but I can't deny the possibility either. I have to answer
>that I need to get a better understanding of what aspects are relevant
>in understanding probabilities.

I do a common mathematical trick: because I know that the "quantifying
probability/credibility" problem is huge, I just prove that, whatever
quantification you use it must be invariant for the situations involved
(with/without delays between reconstitutions, in real, virtual or just
arithmetical reconstitutions, etc.) *once you postulate comp!*.
We don't need to know the proba for understanding that comp forces them
to remain unchanged for those different situations. This is capital for
the sequel of the proof. It explains why we shall be obliged to take the
whole UD* into account. It is due to what I call the invariance lemma
(without doubt with some pompouslyness but it fixes the idea).

>Likewise in question 10, where we assume a UDA is actually implemented
>in the physical universe, I believe it is possible that the measure
>of the implementations in the UDA may be smaller than the measure of
>biological physical implementations.

And what is a biological implementations? And then, how could they
have bigger measure? Remember the UD generates also QUD (Quantum
universal dovetailer). Perhaps the QUD (or "a QUD") wins in the limit
but with comp this still *must* be justified. Don't fall in Searle's
magic naturalism! By definition of comp: if that "biologicalness" is
needed, AND if that biologicalness is non DU accessible, it just means
comp is false. Unless that biologicalness, like the temporalness, or
the space-fillingness are extracted for *first person*, inside or
internal view, of the machines (DU generated) betting on their most
probable computations. Those machines anticipates their own consistency
first. (This could help for the translation UDA -> AUDA).

>So while the UDA machine will
>produce copies of my mental state, it might turn out that they make
>an insignificant contribution in terms of first-person probability,
>compared to the contributions from the physical universe. (For the
>purposes of question 10, we assume that the physical universe exists.)

But in *that* basic physical universe, with comp, the UD will provably
generate continua of variant of that basic physical universe and portion
of it corresponding to your actual states. All our consistent computational
continuations (extensions) count, "biological" (whatever that means) or not.

Hoping this makes clearer my previous comments about your summary of your
post. It seems that your summary points of your last post were
less "magically-naturalistic". You can thrust comp *and* nature in the
sense that [if nature does not comes from comp, then comp is false!].
Variety of "infinitesimal discretisations" could defined weakening
of comp, in that case, making our level of substitution inaccessible.

(In fact nature could perhaps be seen as a sort of consciousness' sharable
border, whith consciousness = stable self-consistency anticipation by
(collection?) of machines.

Received on Sat Jul 27 2002 - 09:22:48 PDT

This archive was generated by hypermail 2.3.0 : Fri Feb 16 2018 - 13:20:07 PST