So I’ve read Maudlin’s argument and I’ve decided to trade in my PC for
a brand new Olympia. Or maybe not – I heard the water bills are a
killer. Actually I’m not so impressed, and here’s why.
Maudlin attempts to show that consciousness cannot supervene on the
physical activity of a computing device by hypothesizing a mechanism
which, instead of responding to counterfactual input states in order
to accomplish a certain computation, is composed of numerous
interlocking mechanisms, each of which is ‘mindless’ in the sense of
producing the same output regardless of what is input (the state of
the ‘tape’ of water troughs), and each of which operates independently
of all others.
In order for this process to work out, Maudlin must ensure that the
trough containing the next bit of information being processed by the
armature is always the next trough being read, i.e., the spatial
sequence of troughs is the same as the computational sequence of
program pi. What then if the armature needs to read an earlier state
of the tape? Rather than allowing the armature to backtrack, Maudlin
invokes the notion of the ‘bilocation’ of certain addresses. By
connecting pipes to earlier troughs, the armature can read an earlier
address in the sequence. This is important because as the machine
progresses through its sequence of operations, it may fill or drain
the troughs it passes over, and these alterations to the tape may be
significant for its subsequent processing.
Now Maudlin claims that each copy of Klara cannot be conscious,
because a machine that produces the same outputs regardless of inputs
can never, by the computationalist’s own standards, be regarded as
conscious. It is only the conjunction of all the various Klaras in
Olympia that achieves a real computation, and yet, because only one
copy of Klara is running at any time, the other copies remaining
completely inert and separate, the computationalist has a problem.
Either he must give up the idea that consciousness depends on a ‘non-
trivial’ computation, or he must give up supervenience on the
Let us note that in order for Olympia to run a particular computation
on a particular input state, she must first be pre-programmed to that
input through the configuration of the float devices at each trough,
which ensure that a new Klara can be activated at each potential
counterfactual branch. This is fine so far as it goes, in that this
can be done in advance and amounts to the programming of any computer
before it is run. But note that at this programming stage, the results
of Olympia’s processing cannot be known, otherwise the program would
have to be run first, and that would lead to an infinite regress. That
is to say, when I program the float positions, I can’t know how
Olympia will fill or empty any particular trough in advance.
So here’s the rub. Let us imagine that Klara instance 1 (we’ll just
call her K1) is running and hits her first counterfactual at T4
(trough 4). K2 is now activated at T4 and K1 is shut down. Now let’s
say two troughs down at T6, program pi calls for the retrieval of the
contents of T1. By bilocation, we know she can do this. But now the
armature has a problem. Should it get the contents of T1 in K1 or in
K2? Clearly the answer is K1, because the armature has passed over K1-
T1 and therefore possibly altered its contents, but K2-T1 has never
been processed and contains ‘noise’ as far as the calculation is
concerned. As noted earlier, we cannot so arrange K2 that it contains
the same contents as K1 after processing, because that would entail
running the program in advance. So K2 must pipe the contents from
trough K1-T1, thereby activating a supposedly inactive copy of Klara.
This itself seems to shoot down Maudlin’s argument, since the
transferring of information (water, charge, whatever) from K1-T1 must
be accomplished physically, and involves the physical, causal
interaction of two different copies of Klara.
The problem is even deeper than this, however. How does the system
‘know’ when two locations should be bilocated? This works OK for a
single copy of Klara, since she is a static system. But if she must
physically interact with all the previous editions of herself further
back in the calculation chain, then she will be forced to ‘build’
pipes on the go, a ridiculously contrived procedure that totally
vitiates the idea of a mindlessly proceeding, inert system. And how
does Klara (or rather, Olympia) remember which path she has followed
in order to know which trough to drain? New mechanisms must be devised
which effectively mean retaining the activity of previous Klaras in
the chain and are no different from a form of backtracking.
If Maudlin’s argument is a foundation of the UDA, then it seems to me
the UDA is on shaky ground, though I have yet to investigate the MGA
in depth. People talk about the Movie Graph Argument, but the links
provided refer to Alice and a distant supernova with lucky rays that
substitute for functional neurons. I don’t see a connection to the
idea of a recording or a filmed graph. Can someone enlighten me?
Bruno, I'm aware I've left a prior discussion hanging regarding
measures of infinite sets. I read up on the material you linked to and
it seemed to me that amenability of integers depends on them being
taken as a sequence, not a mere set. But I don't really (at all)
understand the difference between sigma and simple additivity. Perhaps
you can explain? And is this discussion related to the 'credibility
measure' you mention elsewhere?
On Dec 17, 3:06 pm, Russell Standish <li...@hpcoders.com.au> wrote:
> On Fri, Dec 16, 2011 at 05:42:19PM +0100, Bruno Marchal wrote:
> > On 16 Dec 2011, at 10:39, Russell Standish wrote:
> > >On Mon, Dec 12, 2011 at 04:11:54PM +0100, Bruno Marchal wrote:
> > >>>Maudlin's argument relies on the absurdity the the presence or
> > >>>absence
> > >>>of inert parts bears on whether something is consious. This
> > >>>absurdity
> > >>>only works in a single universe setting, however. If your
> > >>>computer is
> > >>>embedded in a Multiverse, the absurdity vanishes, because
> > >>>thiose inert
> > >>>parts are no longer inert.
> > >>But they do not play a part in the computation, at the correct
> > >>substitution level.
> > >They certainly look like they are. If these parts weren't present, the
> > >calculation proceeds differently in the other branches of the
> > >Multiverse. In other words, counterfactuals are not handled correctly.
> > If you think of a quantum multiverse, then that argument would work
> > if the brain is a quantum computer. If it is classical, its states
> Nobody here is proposing that the brain is a quantum computer. Penrose
> and Lockwood do, but that's an entirely different hypothesis.
> > can be considered as having been prepared in the classical base, and
> > the computation (or non computation) will be handled correctly in
> > each branch of the quantum multiverse, in which the same MGA
> > reasoning will apply.
> In Maudlin's argument, the inert parts are only inert by virtue of the
> counterfactuals not having been realised. In a multiverse, the
> counterfactuals are realised, but in different branches. Hence those
> "inert" parts are no longer inert in all branches. If they were, they
> could be removed from the computer altogether, without affecting the
> > So you are introducing a different kind of physical multiverse,
> > which would handle the counterfactuals. But this will not work.
> > Either this physical multiverse, which plays the role of the
> > generalized brain, is Turing emulable, in which case I can emulate
> > it in a single Turing machine, for which the MGA will apply again.
> > Or it is not Turing emulable, but then the need of it will
> > contradict the comp assumption.
> This step, as I understand it, is a form of dovetailing. Nobody really
> thinks of the dovetailer algorithm as instantiating consciousness, so
> the move is ultimately invalid, I would think.
> > >>They are playing a part concerning the first person indeterminacy,
> > >>like in the UD*, or in QM physics. But that is derived (and has to
> > >>be) from the indeterminacy.
> > >They do that as well, but this is not relevant to Maudlins argument...
> > The parallel realities does not play any role for a classical
> > computation, except for statistical interference (in case of a
> > quantum computer).
> It is not a question of the parallel realities playing a role in the
> computation, but in the supervenience. Maudlin's argument says If
> COMP, then supervenience on single universe is contradictory. But it
> doesn't say anything about supervenience on multiple parallel realities.
> > But if this play a role, it means that we have
> > not chosen the right level of substitution. Once it has be chosen
> > correctly (or below), what happens in some other branch cannot
> > interfere or play any role in the computation.
> I don't follow...
> > >>>If you then fold the multiverse back into a
> > >>>single universe by dovetailing, one can then reapply the Maudlin
> > >>>move.
> > >>Indeed. That is the key point.
> > >>>But then, in that case, one can embed that result into a
> > >>>Multiverse, and the cycle repeats.
> I think I'm coming around to the view that neither of the above steps
> are valid - but one could equally say they are as valid as each other.
> ... snip ...
> > If the register "323" is missing in one branch of a quantum
> > multiverse, it is missing in all normal extension of the
> > computational state of the machine.
> > Some rare branch will have the
> > pieces, and from there (and thus from the first person point of view
> > of the subject) everything will go well, by comp.
> This is a bit confused. Surely the register is missing in all future branches,
> > But only because
> > we fall back in a branch where the piece is not missing.
> Why? Are you saying that if consciousness requires the presence of
> register 323 at some particular point, then we find ourselves
> instantiated by a computer with such a register? But then surely,
> never at any point would we find ourselves instantiated by a machine
> without register 323 - presumably for most of our history we would be
> unaware of whether the register existed or not.
> > This is not
> > different than the comp or quantum immortality argument. The fact
> > remains: the physical activity in one normal branch missing the
> > register is the same as the physical activity in some branch not
> > missing it, for the same particular computation.
> In all branches, or just special ones? If all branches, then the
> register is totally unnecessary. If just a special pair of branches,
> then Maudlin's argument shows that supervenience must occur across
> more branches than those two.
> > Then Maudlins
> > argument shows correctly that the physical activity can be made
> > arbitrary (and even non existing), showing that comp links
> > consciousness not on the physical activity of the program, but on
> > the computational (in the sense of computer science) structure only,
> > making matter and physics an epistemological indexical for the
> > conscious entity involved.
> > >>>The question is - where is the consciousness in all this? I think it
> > >>>must move with the levels - and given the UDA and COMP, I would say
> > >>>that consciousness appears at the Multiverse level, not the single
> > >>>universe level.
> > >>That is right, but with comp that "multiverse" is the mathematical
> > >>structure which needs to be entirely derived from the theory of
> > >>consciousness or from the self-reference logics.
> > >Why? I can see how, but why?
> > Keeping comp, we might say "only by Occam", but that would be weak,
> > given the fact that not much of known physics is handled by comp
> > currently.
> > But then the reason why we have to do that, even without Occam, is
> > the MGA argument. If some physical reality is at play in the brain
> > for it having a role in the making of consciousness, comp makes it
> > Turing emulable in a single reality, and it that single reality we
> > can change the computer structure so that his physical activity is
> > arbitrary, by adding, like Maudlin some physically inactive piece of
> > matter, for handling the counterfactuals. And what I say above will
> > apply.
> Yes - it shows that physical supervenience is impossible in that
> single reality.
> But this doesn't answer the why question. I could imagine that you
> might feel that Multiverses are otiose, so would prefer a derivation
> of their existence from something "simpler" - eg arithmetic of the
> whole numbers.
> That's fine and dandy - but the Multiverse is not otiose - it is far
> less of an impost than a single reality.
> I know you're keen to attack the Aristotelian primary matter
> position. To be quite honest, its not a question I care too much about
> - I'm happy for my matter to be phenomenal, not primary. But I do
> think we need to be careful about throwing out supervenience of mind
> on matter (of whatever stripe), otherwise the Anthropic principle
> becomes mysterious, and we're faced with what to do about the Occam
> I'm snipping the following text because it moves away from Maudlin's
> argument in particular, and also there's some juicy stuff in it I need
> to absorb before responding (if indeed I do :).
> Prof Russell Standish Phone 0425 253119 (mobile)
> Principal, High Performance Coders
> Visiting Professor of Mathematics hpco...@hpcoders.com.au
> University of New South Wales http://www.hpcoders.com.au
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To post to this group, send email to email@example.com.
To unsubscribe from this group, send email to
For more options, visit this group at