[ExI] Do digital computers feel was Re: Is the wave function real?

Rafal Smigrodzki rafal.smigrodzki at gmail.com
Sun Dec 11 06:26:23 UTC 2016


On Mon, Dec 5, 2016 at 12:38 AM, Colin Hales <col.hales at gmail.com> wrote:
>
>
> Whatever the universe is made of,  you and I and the rest of it are all
> made of it.
> Whatever the universe is made of, it is capable of creating an observer
> (scientists) inside it.
>

### It occurred to me that "conscious" observers (i.e. experiencing qualia)
could be qualitatively different from digital simulations of minds, even if
their behavior was indistinguishable. Of course we do not know much about
the subject, and have to use intuition rather than more explicit knowledge.
Let's then examine our intuitions and try to identify boundary conditions
under which they might be true.

Let's consider a neural network, capable of reading the alphabet, being
shown pictures of a letter. Furthermore, this network is built from a very
high resolution scan of a human who reads. The fidelity of the simulation
is arbitrarily high but it is a digital simulation - one not using quantum
phenomena or other forms of analog computation.

Let's run this network a hundred times from one starting point, feeding it
the same input, for example the letter A. Obviously, it will produce the
same output - "I see the letter A". Do we have a human mind there, reading
the letter a hundred times?

I don't know. The run in a digital computer consists of operations on
bitstrings - input, processing, output, all of that can be described by a
single finite bitstring. Each of the runs is equivalent to a re-examination
of the same bitstring. If the identity of indiscernibles holds, then all
the runs are in fact analyzing a single mathematical object. The same
pertains to runs simulating a human hit with a baseball bat - all the runs
are equivalent to a single bitstring. Is there a human screaming a hundred
times in agony in your computer, or is he screaming once, or is it just
unfeeling numbers, a digital ghost devoid of qualia?

I find it difficult to think that a single mathematical object can
experience qualia every time it is examined with the use of a digital
computer.

Let's continue our experiments. The human simulation will reliably read "A"
even if you bit-flip a single simulated molecule in a simulated synapse -
the way the human brain is built, you cannot affect a high level quale,
such as perceiving a letter, by introducing a small amount of noise below
some threshold. The simulated digital mind will be similarly resistant to
digital noise introduced at that level. In fact, you could bit-flip every
single synapse in the mind, which is equivalent to a brief epileptic
seizure, and still get the same reading, slightly delayed. You could
generate a large number of specific, non-identical instances of this
digital noise and use them to run the network a hundred times. Each of the
runs would be the equivalent to a different bitstring, a hundred different
mathematical objects. Do we have a human reading "A" or being hit with a
bat a hundred times here, or not?

I don't know. If all you need to generate consciousness is to calculate a
bitstring according to a digital program then yes, there are a hundred
instances of a feeling human here. But there is something off here. In the
experiment without digital noise, identity of indiscernibles makes it
difficult for me to claim there were multiple separate instances of
consciousness. How could a single bit flip, completely incapable of
changing the reported experience ("I see the letter A"), manage to generate
separate instances of consciousness in the experimental runs?

My guess here is that digital noise alone cannot trigger separate instances
of perceiving qualia in digital computational runs. Such experiments look
at the properties of idealized, platonic forms, built of finite numbers of
elements. I guess that finite mathematical objects are qualitatively
different from our physical reality. On a metaphysical level, I would
surmise that observable physics contains mathematical infinities. This is
of course just an idle act of faith on my part, not any well-formed type of
knowledge. I don't understand infinite-dimensional Hilbert spaces, so my
hunch about our QM world being built of infinities is just
pattern-matching, not inference.

Let's continue our experiments. We download out simulation into a hundred
physical copies of the human brain. Every neuron, every synapse and every
atom is in the same corresponding spot in all copies. We show them the
letter A by stimulating their optic nerves and all produce the output "I
see the letter A". Using our advanced monitoring hardware we verify that
the physical noise at the level of individual neurons is similar to the
digital noise introduced by out bit flip in the previous experiment. Do we
have a hundred humans here thinking separate conscious thoughts?

Most of us would say yes. A hundred physically existing human brains,
generating the known observable correlates of thinking - EEG patterns,
synapse activations, producing verbal output - that's a hundred conscious
thoughts experiencing qualia, or else no human is conscious.

We have three sets of experiments on similar entities yielding different
intuitions about the entities. What has changed between the sets?

As I hinted at above, first two sets are reducible to finite bitstrings
(i.e. can be completely recorded and predicted by manipulation or
properties of finite bitstrings). Whether the third set can be described by
finite bitstrings would depend on the ultimate physical nature of reality.
If quantum-mechanical physical objects (a bit of a pleonasm, isn't it) are
indeed defined by mathematical objects that are not equivalent to finite
bitstrings, such as real numbers, uncomputable numbers, or others, then an
analog computer such as a human brain or quantum computer would be
qualitatively different from a digital computer, in an irreducible way.

For any human brain you could produce a digital simulation that over at
least some time span would behave identically to the physical
instantiation, and even retain an identical high-level information
processing structure. If there is an irreducible difference in the
mathematical objects underpinning the analog/human and the digital
simulation, it might be possible for metaphysical zombies to exist.

My argument is not constructed to support dualism in the usual sense, it is
not a critique of physicalism, and it goes off at a tangent to the zombie
literature I am familiar with. Rather, I am trying to organize my thinking
about modal realism. I am definitely a physicalist when it comes to the
hard problem of consciousness but then I am also a modal realist when it
comes to metaphysics. I do believe that physical existence is a member of
the ultimate ensemble of mathematics but then our physical existence may
use more complicated mathematics than can fit in our minds. Or our digital
computers.

I intuit that digital objects exist outside of time, isolated from our
time-like reality, like some other platonic mathematical forms. We do not
interact with them in the same way as we do with physical objects, we
merely examine their existence. But we interact with the mathematical
objects that make up our physical reality, flowing through infinities of
time-like correlated states, and our consciousness is how higher
mathematics feels from the inside. I am modestly inclined to believe it but
I am not quite sure. Maybe merely embedding a model of a platonic
mathematical object in our physical computers imbues such object with new
properties? The mind boggles.

On a more practical level, if there are digital p-zombies, it means it's
morally OK to shoot NPCs in digital computer games, no matter how life-like
are their pleadings. Or perhaps my intuition is wrong, and little programs
running reinforcement learning could be tormented souls.

Rafal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20161211/e5b135be/attachment.html>


More information about the extropy-chat mailing list