[extropy-chat] Fragmentation of computations

Eugen Leitl eugen at leitl.org
Wed Mar 21 13:18:27 UTC 2007

On Wed, Mar 21, 2007 at 11:39:04PM +1100, Stathis Papaioannou wrote:

>      I don't know what computationalism is, so I take that as a
>      definition.
>    It's not the definition, it's what I'm trying to prove.

It's an interesting problem. I don't think it makes much sense, though.

>    Computationalism is the idea that you could swap your brain for an
>    appropriately configured digital computer and continue to have the
>    same kinds of subjective experiences. Computationalism would be wrong
>    if we had a God-given soul or if Penrose were right and the brain is
>    not Turing-emulable.

I can't rule out computationalism (what an awful word) is correct, but
that's my working model. It would be very difficult to patch up a live
system, especially a formerly alive system which is a cryogenic chunk
of tissue glass.
>      Computationalism means that the trajectory sequence doesn't matter.
>    I think it does, but it appears that you think it doesn't.

I'm seeing big practical problems with shortcuts. The world is nonlinear,
and the CNS is especially nonlinear, so there can't be any deep shortcuts.
You're limited to resuming from snapshots. That's not out of sequence,
that's strictly sequential. If the computation is completely reversible,
then it would be also interesting to see subjectively what happens
when you run it in reverse.

Apart from that, I don't see many other opportunities to game the system.
>      The immediate objection: which process produces the trajectory
>      slices
>      which are out of sequence, but belong to the same trajectory?
>      Enumeration only works for trivial state space sizes. Computing
>      shortcuts are available only for some trivial systems, so you have
>      to compute the trajectory in sequence, which defies the purpose
>      of assembling the trajectory out of sequence.
>    That objection relates to how the states will be generated, not to
>    what would happen if they were generated out of sequence. If I asked

I have reason to suspect that they can't be generated out of sequence.
It's like speculating what would happen if I could jump to the Moon
straight out of my chair. Simple physics tells us we can't, so we
can spare the speculation.

>    you whether a car would still function as a car if it were thrown
>    together in exactly the right configuration by a storm rather than
>    assembled in a factory, would you answer that it wouldn't, because it
>    is very unlikely that such a thing would happen? In any case, you can

Precisely. This is not something which can't happen in theory, it can't
happen in practice. Stochastical processes don't build up complexity
without some secret sauce guidance.

>    assume that the computation is run in sequence the first time, then
>    out of sequence on the second run using information from the first
>    run.

How exactly would you run a simulation out of sequence? Here's a gas box.
It computes system refreshes (trajectory frames) once a second. It
dumps a snapshot image every minute. You don't know what the snapshot
image after an hour is, without running the computation.

You can go back to the snapshot frames, and and resume from there,
but it doesn't change the end result. But you can't shortcut to the
end result without doing the work.
>      How do you verify the subjective experience? The only way to do
>      that would to
>      access the trajectory frames, which requires information about
>      the sequence of trajectory frames.
>      See, there's a giant can of worms implied.
>    You verify it the way any psychological experimenter would: by
>    observing the subject's behaviour, in this case perhaps on a computer
>    screen if the program includes a visual representation of the action.

But you're running things out of sequence (which is usually impossible,
but Hash Life gives you a minor leeway here). How can you measure
behaviour which is out of sequence? You have to access it in sequence,
so you're doing the work. Not the system.
>      >    computational states as well as on the present instantaneous
>      state,
>      >    then arbitrarily slicing up the computation will change and
>      perhaps
>      How would you "arbitrarily slice up the computation"?
>    By stopping it and starting it at various intervals,  changing the
>    clock speed, changing the computer, saving it at various points and
>    using this information the second time to run it out of sequence.

How can you run a linear computation out of sequence? You can run it
backwards, assuming it's reversible (most are not). You can resume
from snapshots, but this is all forwards or backwards, each slice
precisly in sequence. 

Show me how you would run things out of sequence. In a generic
case, not in Hash Life sense. 
>      You keep using that word. Nobody but you knows that that word
>      means.
>    Consciousness, supervenes on or instantaneous?
>    consciousness =  having experiences

Having experiences means as little to me as consciousness (i.e. I can't
work with that definitions computationally, they're too diffuse), but
I'll agree with that for the moment.

>    supervenes on = a little involved, see
>    [2]http://en.wikipedia.org/wiki/Supervene, but let's just say it means
>    "caused by" for the sake of this argument

It is so involved that it's not saying me anything at all.
What is supervienience good for? What can I build with it,
what I can't without it?

>    instantaneous = same usage as in calculus (approximately, since
>    computational states are discrete)

I don't see where you refer to instanteous in your reasoning, but
instanteous doesn't exist. Not in theory, in practice. Computation
takes spatial extent, which is equivalent with time in a relativistic
>      >    computational state (or more simply, consciousness *is* the
>      >    instantaneous computational state), which makes it impossible
>      to know
>      >    from the inside whether your computation has been fragmented.
>      You still haven't replied how to verify the subjective experience
>      of a
>      Hash Life observer in a observer/virtual world implemented in the
>      Life
>      CA. Verify as: measure.
>    You can't really *know* what someone else is thinking, but you can
>    observe his behaviour or ask him.

This is precisely what I'm asking you. How can you measure the
behaviour in a sim which is out of sequence? Specifically, in a
Hash Life world? It is the only kind of world which allows nonlinear
spacetime computation.

Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
ICBM: 48.07100, 11.36820            http://www.ativel.com
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE

More information about the extropy-chat mailing list