I'd like to expose some arguments against computationalism - mainly
elaborated through the various discussions on this list, many thanks to all
contributors!
I think we should distinguish between two forms of computationalism, which
are unfortunately often confused in many discussions:
comp1 : the brain is an EXACT implementation of some digital computation
equivalent to a TM. The running of this computation on any equivalent TM
would produce exactly the same output.
In other words, we are just running a very complicated digital program
(much like Word but still more complicated!) that could run on any platform
(complicated enough). Like our ordinary computers, the implementation must
be independant of the material structure of the platform, and thus the
information must lie at a (much) higher level than the molecular one, most
probably at the level of a neuron state, just like the bits of a computer.
I think we will all agree that the experimental facts from neurobiology are
enough to conclude that comp1 is FALSE, because a neuron state can not be
defined in a discrete, deterministic way. The firing pattern depends also
on many analogic quantities (temperature, concentration of various
neurotransmettors, possible drugs) and has a strong probabilistic
(pseudo-random) behaviour.
Most of you seem to conclude : "ok, no problem. We know that all these
analogic parameters can be modelized by deterministic equations; it's
enough to implement the solution of these equations to produce the same
output as the brain, and THUS consciousness, and that's it.".
Most of the (if not all) devices dicussed as "artificial conscious
machines",(including Bruno's "crackpot" dreaming machine) are based on this
assumption. As the brain is in fact interacting with the outer world (a
point that would deserve more discussions : I think that even dreams or
hallucinations would not be possible if the brain had NEVER interacted with
its environment), you soon realize that the equations must take into
account the whole Universe, a la Schmidhuber. To recall last Tegmark's
quote:
> Let us imagine a hypothetical Universe much larger than our own,
> which contains a computer so powerful that it can simulate the time-
> evolution of our entire Universe. BY HYPOTHESIS (*I emphasize*), the
>humans in this
> simulated world would perceive their world as being as real as we
> perceive ours, so by definition, the simulated universe would have
> PE [physical existence].
So you are led to another hypothesis, that I will call comp2, which is in
fact Bruno's "comp" postulate.
comp2 : all the physical world is EXACTLY equivalent to some computation of
its analogic properties at a finite level .
As Bruno shows it magnificently, comp2 leads inevitably to the actual
disappearance of the physical world, replaced by a world of computations.
However, I want to stress here that comp2 is indeed a huge jump: in fact it
does not COMPLEMENT but rather CONTRADICT comp1; it is the source of all
paradoxes and worst, it it completely useless to understand the origin of
consciousness.
comp2 contradicts comp1 because the essence of comp1 is the independance of
the results with respect to the material implementation, whereas comp2
requires the precize definition of this implementation (of course the
simulation made in comp2 could be runned on any TM, but the object of the
simulation must be one precise physical system).
One of the problem is to find the actual level at which this simulation
should be made. As I already stressed, the physical laws we are using do
not describe the REALITY, but only our REPRESENTATION of it. Although it is
plausible that there is SOMETHING objective, we have no idea what this
could be ("das Ding an sich following Kant). So which representation to
choose ? The chemical description ? The QM state ? String theory ?(with
the supplementary difficulty that beyond the QM level, no measurement of
the Q-state is possible). You could try to accept any level giving an
acceptable output, but then isn't Hans right when he assumes that even
Teddy bears and movies characters have "acceptable enough" outputs to be
considered as conscious?
Note that this first difficulty did not exist with comp1, but we are facing
it inevitably with comp2.
Assume you have solved this difficulty, either by finding a known (e.g.
electro/chemical) level at which the brain evolution IS predictible, or by
finding a (subquantum) TOE reproducing exactly all know features of the
Universe. You may hope to describe your brain at this level and calculate
its evolution. You may think (with comp2) that a TM calculating the state
of your brain would actually be conscious LIKE YOU.
I will not recall all paradoxes associated with this hypothesis (for
example Olympia/Karas paradox) But taking again just the chinese room
example, let think of the case where your brain would be actually simulated
not by a machine but BY SOMEBODY ELSE ? Who or what would feel your
consciousness ? You could imagine a situation where the entirely state of
your brain at some instant is stored in a huge library, and somebody (for
example me) is put in charge to calculate its evolution (for example
applying id\psi/dt = H \psi to a quantum state). What do I have to do for
this device to actually think (like you, not me) ? Must I write the output
of my calculation somewhere? with a pen? on a magnetic tape? what if I read
a stored file of a previous calculation ? And if it is read by somebody
else that does not know what it represents? Please tell me ! I am paid for
this job and I don't want to be fired!
Of course I just point out again the contradictions between comp2 and
physical supervenience, but abandonning Phys-sup does not solve the first
point. It makes it still worse because abandoning the idea of physical
reality (Marchal) means also the impossibility of linking a computation
with a physical state, which is the starting point of comp2 and
Tegmark-Schmidhuber and co..theory!!!
A third and last difficulty is that comp2 does not solve the problem of
consciousness. For if everything is assimilable to computations, what makes
some computations or parts of computations conscious or not (see Wei)? So
how to found Bruno's "computational psychology" ? What is the dream of a
string? Complexity is not enough, because for example the chemical
evolution of a thinking brain is not more complex to that of a dead brain
leading to putrefaction. At the analogic level of comp2, you have lost the
information level of comp1 ! I agree that the problem is the same with
materialism - I just point out that it is not easier with comp2.
So I think that pure computationalism, either comp1 or comp2, is very hard
to maintain. Another comp3 proposition?
One remark is that all "thinking" devices based on digital simulations of
the analogic state of the brain handle in fact much more (and too much)
information than the brain itself, which is totally unaware of its own
material structure. A very important fact is that they ALL require an
external structure able to store the relevant information and program them
adequately, which is NOT the case of actual brains (and I guess of possible
future thinking machines).
So my guess is that consciousness requires not only a proper handling of
information, but that this handling must be a natural consequence of
physical evolution without ANY interaction with an external storage of
information about its own structure, even for its construction. This would
insure the proper handling of counterfactuals which is in fact nothing else
than the self-construction of consciousness. This is NOT to be taken as a
formal definition of consciousness but only a possible restriction to get
an actually conscious device. It would not be pure computationalism because
it would put restrictions on the kind of implementations it requires.
Comments?
Gilles
Received on Sun Aug 01 1999 - 07:36:41 PDT
This archive was generated by hypermail 2.3.0
: Fri Feb 16 2018 - 13:20:06 PST