On 23 Apr 1998 mod...@concentric.net wrote:

BJ: And may devolve upon a more *complete* physics.

Neural nets provide something of

BJ: Mere complexity is not the critical issue.

BJ: Right.

  Architecture affects practical issues of

BJ: No, this is only a silly dogma spawned by AI types.

BJ: More of same. The architecture has crucially to do with what kinds of
sensory input can be operated upon.

On 23 Apr 1998 mod...@concentric.net wrote:

BJ: And you have determined this ... how?

BJ: What principle are you invoking?

BJ: Wonderful finality, that--but perhaps it is only a question of what
one means by 'computer'.

http://www.freeyellow.com/members/sentek/index.html

In <Pine.SV4.3.91.980424064603.647B-100...@sleepy.giant.net>, Brian J Flanagan <bflan...@sleepy.giant.net> writes:

It's not an "AI" idea.  It's one of those things that is basic to the
notion of computing.  A guy called Church formalized it quite a while
ago, but it is pretty obvious for discrete digital computing (once
you've realized that no discrete architecture can transcend a Turing
Machine), and more subtly obvious for all machines of any type.

I don't know how you mean this.  

It could be the same point I made in the paragraph you quoted and agreed
with above...  that hardware is needed to provide sensory input to (and
effector output from) a computer system.  It's true in that sense, that
a computer can't operate on input that it doesn't have.  But then
why post it as a disagreement?

It could be an assertion about practical performance issues.  We
probably do in practice need some highly parallel architecture to keep
up with the sheer volume of data involved in a realistic modelling of a
world reported through a full sensory interface, and architectures which
implement primitives directly related to the functions we need are
faster than those which have to build up to them from a different set.

But if you really mean to say that the architecture used for the
computing itself makes a difference to what can be computed, given the
necessary input and ignoring performance...  then I respectfully suggest
that's incorrect.

"Computing" is generating functions as combinations of other functions
given as primitives.  Any computing architecture capable of a few very
basic operations can compute the primitives of any other, and thus can
go on to compute any function computable by the other.

At least according to Church and me.  <g>

Bill Modlin

The important points that you are missing are:

        We are not given the necessary output.  We have to fetch our
        own input, and make our own decisions as to what input to
        use.

        We have to make do with whatever performance we have.  It it
        took a year to make the decision whether to eat that morsel
        of food, we should soon starve to death.

Um, not an expert on this myself, but...

Isn't there a mathematical proof by someone that a syntactically derived
system is capable only of certain things, and is inevitably stumped by a
particular set of problems when confronted with them?

The Chinese room, thing.

You know, when an englishman is replying to chinese questions according
to a set of books that define how to answer, as each question is put
under the door, the englishman refers to a book which tells him the
answer to give back. Can this englishman be said to understand chinese?

I'll take a stab at this. In principle, my elderly computer was
perfectly capable of producing the same answer to a given question as my
spanking new PII, except that it takes quite a lot longer to get to the
answer.

This is not a completely satisfactory answer, because win95 checks to
see what cpu I've got and won't run on an 8088, but the principle
appears reasonable.

Ones which use syntax and serial processing seem to be poor candidates
for an AI computer. Imitating intelligence, maybe.

And if a rock can be sentient, it would be better to let the computer
decide for itself what to think, rather than putting in a program which
permits no thoughts except those written in stone and coerced using
error correction. Else, what you see is what the programmer told it to
say, not what it is saying. This changes but little if the program can
evolve, it's still a program rather than AI.

In <6hqde2$...@ux.cs.niu.edu>, rick...@cs.niu.edu (Neil Rickert) writes:

I'm not missing those points.  I'm explicitly talking about computation,
to which they are irrelevant.  They may have a lot to do with whether a
computation is useful or effective, and we need to consider them in
talking about what is needed for consciousness and intelligence...  but
they have NOTHING at all to do with whether different architectures can
compute different functions.

Computation is transforming data according to some functional
relationship.  What we call a computing architecture is a set of
primitive functions plus some means of combining them to make up other
functions not defined as primitives in the architecture.  It turns out
that any of many very simple sets of primitives is enough to allow
combinations implementing any other computable function.  We call an
architecture capable of at least such a set of primitives "Turing
complete", and any such machine can compute any function any other such
machine can compute,  given enough resources.   This is analogous to the
notion of a "Boolean complete" set of primitive boolean operators such
as (AND OR NOT) or (NAND).  Given a boolean-complete set of boolean
operators you can generate all possible boolean functions, and given a
Turing-complete set of computing primitives, you can compute all
possible computable functions.

I'm finding it frustrating that you keep posting that you disagree, when
I know that you understand this point because you've made it clearly
yourself, several times.  Why disagree when I say the same thing?

Bill Modlin

Fair enough.

In that case I am now in a position to assert definitively that
computation, as you are using the term, is completely irrelevant to
cognition and to intelligence.

I once would have thought so.  But that guy Modlin has just persuaded
me otherwise.  He has persuaded me that 'computation' is a pointless
game of mechanically manipulating meaningless symbols according to a
completely arbitrary set of pointless rules.  I will have to take his
word for it.  After all, what right do I have to argue as to what is
computation?  I'm only a mathematician.

That sounds more like the definition of a function algebra.

Perhaps the reason I keep posting that I disagree, is that I
disagree.

In <6hqj4a$...@ux.cs.niu.edu>, rick...@cs.niu.edu (Neil Rickert) writes:

  [responding to my attempts to distinguish "computation" from the
   input/output activities which supply data for computation]

Ok Neil.  Please.  What is it that you disagree with?  
What is your definition of computation?

Under your rules, is a Turing machine capable of computation?

Under your definitions,  if I load a program and some data into a PC,
and start the program running, can the PC compute some function of that
data without using any further I/O?

I do understand that you consider interfaces with the outside world to
be much more important to cognition than any computation which may be
involved.  But I also thought that you agreed that somehow there was
some computation involved in figuring out what to do with the data you
get through sensors, and deciding what other information to go looking
for, and all that sort of stuff.

I've seen you tell others that it didn't matter whether the interior
processing was done with neural nets or an analog computer or a digital
computer, that all of them could do the same things.

When I say just that, you tell me you disagree, and get sarcastic about
it.  Why?

Can you explain?  Help me out here?

Bill Modlin

A computation is a set of causal operation which take place in the
world, and which have a certain kind of mathematical description.

No.  It is capable of formal computation, but not of computation,
where formal computation is a mathematical idealization of
computation.

It is worth pointing out here that mathematicians carefully
distinguish between pure mathematics and applied mathematics.  Most
mathematicians particularly value pure mathematics, and I count
myself amongst that 'most'.  Generally, pure mathematics comes higher
in the pecking order than applied mathematics.

The Turing machine has always been considered part of pure
mathematics.

Computation has always been considered something done in applied
mathematics.

That's a tricky question.  Even though there is no further I/O, there
are still causal operations going on in the machine, and presumably
these could be probed with suitable measuring instrumentation.  If
you are talking about those causal operations, then you can make a
case that computation is still taking place.

Let's talk about the computer on my desk, instead of cognition,
although the same points apply.

When a computation is being carried out on my computer, a series of
causal operations is taking place, using electrical signals and other
kinds of physical operations.  There is also a symbol manipulating
game used in our theories of computation to describe what is
happening.  Whether we should say that the symbol manipulation game
is actually going on is itself a tricky question.  I prefer to say
that the symbol manipulation game is part of a description language,
rather than part of the computation.

If somebody should come up with a completely different symbol
manipulation game to describe what is happening in my computer, that
would not change what the computer is doing.  The exact same causal
actions would be going on.  And if a computer is sufficiently
complex, it is very likely that there are alternative symbol
manipulation games that could be used to describe the processing.
When I am using the computer for ordinary things (like writing this
message), I care only about the causal operations.  The particular
choice of symbol manipulation game which somebody might choose to in
their description language is completely irrelevant to the
computation that is going on, as far as I am concerned as a user of
the computer.

It seems to me that you are emphasizing only the symbol manipulation
game, and ignoring the causal operations which I consider to be at
the heart of computation.  I prefer to think of the symbol
manipulation game as a story involving purely theoretical entities.
In that sense it is comparable to scientific use of massless charges,
frictionless motion, and other such theoretical entities.  We value
these theoretical entities because of the way we can use them in our
predictive theories.  But we don't actually believe they are more
than convenient fictions.

Sure there is.  But that computation consists of causal operations,
rather than the symbol manipulating game we might use to describe
those operations.

That's not quite what I have said.  Given a particular computation
considered as a sequence of causal operations done in real time, it
does not matter what particular additional machinery was used to
implement those causal operations.  But it does very much matter
whether a particular piece of hardware is capable of carrying out the
required causal operations within the allowable time frame.

On 23 Apr 1998 17:00:42 -0500, rick...@cs.niu.edu (Neil Rickert)
wrote:

Me too

Indeed

I once heard a quite strond argument during some introductory AI
classes: computer hardware (neural nets not included) work in
algoritms. Conscious minds, such as ours, use procedures (or whatever
you want to call it) that are not algoritm based. Computers CAN only
use algoritms (at least nowadays) so based on this principle, a
computer will never gain consciousness, no matter how big or fast it
is.

Greetings.
Wim.

In this  world, all you need is honesty and sincerity.
If you can fake those two, you're set for life.
                    ---- Groucho Marx

I would simply say that computers follow algorhythms and minds cognize
algorhythms and, to the extent an algorythm has been cognized, are
therefore "_free_" to utilize or follow the algorhythm or not.  
I would add, much at my peril in this particular newsgroup, that the
latter function probably also entails "standing outside the system"
(Lucas) or 'being able to "see" what is going on' (Roberts).

--

               Phil Roberts, Jr.

Feelings of Worthlessness and So-Called Cognitive Science
        http://www.geocities.com/Athens/5476

Dennett maintains that 'Turing has proven- and this is probably
his greatest contribution- that his Universal Turing machine can
compute any function that any computer, with any architecture,
can compute'.

This is a common error in rendering the Church-Turing thesis.
A UTM can compute what is computable for any Turing machine.

Thesis M:"Whatever can be calculated by a machine (working on
finite data in accordance with a finite program of instructions)
is Turing-machine-computable.

Thesis M itself admits of two interpretations, according to
whether the phrase 'can be calculated by a machine' is taken
in the narrow sense of 'can be calculated by a machine that
conforms to the physical laws(if not to the resource constants)
of the actual world', or in a wide sense that abstracts from the
issue of whether or not the notional machine in question could
exist in the real world. The narrow version of thesis M is an
empirical proposition whose truth-value is unknown. The wide
version of thesis M is known to be false. Various notional
machines have been described which can calculate functions
that are not Turing-machine-computable(for example, Abramson,
da Costa and Doria, Doyle, Hogarth, Pour-El and Richards,
Scarpellini, Siegelman and Sontag, Stannett, Stewart, Copeland
and Sylvan is a survey)."

So "no discrete architecture can transcend a Turing Machine".
I assume that this alludes to the 'narrow sense' above.
This is a stronger assertion not contained in the CT thesis.
You probably consider it an implication; but this implication
is not proven and does not refute architectural considerations.
I am not sure what "and more subtly obvious for all machines
of any type" means. If this is meant to apply to the 'wider
sense' above, there are counter-examples.

Regards,
Stephen

In <6hqseq$...@ux.cs.niu.edu>, rick...@cs.niu.edu (Neil Rickert) writes:

[modlin] What is your definition of computation?

[modlin] Under your rules, is a Turing machine capable of computation?

Interesting.  Your conflation of "computation" with notions of
physically realized causality is something I've not encountered
before...  none of the classic works on computability uses it that way,
and indeed I can't think of a single author who would balk at saying
that a Turing machine computes.  Computation is an abstraction,
inherently distinct from the engineering practicalities of a device
which might instantiate it.

I think I'll just stop talking to you about this... it would be too
annoyingly clumsy to adopt your idiosyncratic usage just for that
purpose.  Unfortunate.

Bill Modlin

I don't think I am conflating anything.  Many big corporations have
been purchasing expensive computers for decades, because of the causal
operations that they perform.

I would use 'computability' in the same way as the works you refer
to.  You have to remember that mathematics ain't real life.
Mathematicians work with idealized models of real life.  The Turing
machine is an idealized model of computation.  The mathematical
theory of computation is a theory of this idealized model, just as
the mathematics of the real numbers is about an idealization of the
decimal measurements we make.

In a context of talking about the mathematical theory of computation,
I would also have no problem with saying that a Turing machine
computes.  But in that case the context reminds us that we are
actually talking about a mathematical idealization of computation.

Well if that is right, then there is something terribly wrong with
the idea that cognition is computation.  To make that claim, you
would either have to say:

   there is a little man inside having abstract thoughts, and doing
   abstract computation, and that is what creates human cognition;

or

   God is having abstract computational thoughts, and that creates
   human cognition.

In the first case, you haven't explained anything, for you had to
hyupothesize the little man inside.  In the second case you have a
severe case of substance dualism.

In <6hrkup$...@ux.cs.niu.edu>, rick...@cs.niu.edu (Neil Rickert) writes:

(Before responding, let me offer an apology for saying I wouldn't talk
to you about this any more.  I was frustrated.  But now after a few
hours sleep I wish I hadn't said it. <g>)

Anyway.  The "causal operations" which make computers worth money to
a business are outside the computers themselves.  A corporation would
not care if the internal mechanisms of the computer used entirely
different sets of causal operations to accomplish its computations, or
even if it worked by some mystical acausal magic...  so long the
external results have a proper functional relationship to things of
interest to them.

Thats where I see your words as conflating ideas which would be more
usefully separated:  you seem to me to be trying to use the external
mapping of computational results to motivate arguments about the
internal mechanisms of computation.   The two are related only
indirectly, through practical engineering issues of performance.

But when I as a programmer try to design a procedure to accomplish some
useful computation, I'm looking at as a mathematical idealization.  I'm
certainly not concerned with the causal details of NAND gates etched in
silicon.  I work in terms of an idealized machine, an abstract language
virtual "computing device" such as that defined by the C language, and I
seldom care how that abstraction might be physically realized.

The normal working environment for anyone concerned with computation is
an "idealized model of real life", linked only at its periphery with
anything actually real.  The word computation is never used in the sense
you are contrasting with the mathematical, in reference to some process
intimately wedded to its particular realization.

Which is what I meant by the following:

You reacted to the above statement by saying this is incompatible with
the notion that cognition is computation, and doing some handwaving
about it implying an homunculus or substance dualism.   I'll try to
respond to those remarks in a separate posting.  For the moment, I'm
more interested in whether or not we can get past our disagreement about
the relationship between computation and causality, without worrying
about its implications for cognition.

Your turn.

Bill Modlin

What about simulation?  Using algorithms you can simulate the behavior of
the human brain, or of a neural net.  The sophistication is an emergent
quality, not apparent at the finer scale of processing.  The processing
involved would be immense, and the system would have to be far more
sophisticated then the one being simulated (modeled?) (emulated?) to be
effective.  I heard a phrase, "Any computer can emulate any other
computer", I suspect it's true.  Once again though some emulation's would
require far more processing power than than the original computer being
emulated.

Leads me to a question I always wondered about, say we hit the point where
we can actually map a human brain to a fine enough detail that we can
simulate it's behavior on a computer.  Will the simulation be conscious?

Mdg

In article <6hq7pl$...@examiner.concentric.net>,

BJ: What do the elements of your theory of computation correspond to in the
natural world?

BJ: Church the logician, I presume. The formalist stance can be traced back to
antiquity, but this viewpoint has achieved the status of a modern dogma at the
hands of AI enthusiasts. To what do the elements of our experience correspond
in a formal theory (T)?

BJ: Sorry, that wasn't very clear, was it?

BJ: Good!

BJ: How does one compute the color 'red' as experienced?

BJ: Primitives are what we operate upon, yes? I.e., they are the arguments of
our elementary functions?

-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/   Now offering spam-free web-based newsreading

On 24 Apr 1998 mod...@concentric.net wrote:

BJ: Leibniz argued for a similar view, but he has largely been ignored.
Can you carry out a computation in a (natural or artificial) system which
does not respect physical causality? How do you account for Wigner's
'unreasonable efficacy' of mathematics with respect to the 'real',
'physical' world? Is it not perhaps 'because' our theories are abstracted
from that world?

Computation is an abstraction, inherently distinct from the engineering
practicalities of a device > which might instantiate it.

BJ: Can you name a computation which exists apart from a set of hardware
or wetware?

My argument towards equivalence between cognition (or any other
physical process) is emulation.

I.e. if we consider cognition to be a large-scale behaviour of a given
physical structure (ion channels, lipid bilayers, cytoskeleton, etc.),
then by simulating the low-level processes sufficiently accurately I
would be able to reproduce also the high-level behaviour, in this
particular case, cognition.

In fact I am doing exactly this, investigating intricacies of
biomolecules: proteins and lipid bilayers on a vanilla PC, and not
even resorting to quantum weirdness: newtonian paradigm sufficies
here. In fact there are researchers who use compartmental neuron
models with some success, which shows that the molecular level is
actually not relevant -- you can do it at a much higher level.

Before you say that simulation of a thing and the thing itself are not
equivalent: very true, but cognition is one of the cases where
emulation and the process itself are equivalent. A computer simulating
playing chess, _plays chess_, after all.

Regards,
Eugene

ffdsff

Mgd is right!  Simulation is the way to approach AI.

We live in an analog world.  Converting analog to digital samples is not
simulation of the brain activity.  In order to properly program the
interacting partial differential equations necessary to simulate the
instantaneous response of brain functions, it is necessary to use an
analog computer.  Digital inputs can be used to change the function
coefficients, much as the neuron synapses use them.  Every probe in the
brain (by Robert Heath in the '70s) produced recordings similar to EEG
signals: analog traces, not digital.  It has been done in the laboratory
before.

Now all you AI researchers have to do is to develop a stable million
channel analog computer to simulate a cat's consciousness.

Bob

I was not offended by the comment.  Of course you have a right to
decide what you will discuss, although perhaps it is unwise to
publicly declare how you will use that right.

No, I disagree.  They are in the computer.

That is about right, provided that the computation was actually
accomplished, including suitable external causal effects.

I think you don't know much about corporations.

There could be no external results unless there were internal causal
operations.

You seem to have entirely missed a point I made.  Namely, there might
be a completely different way of describing the internal operations
of a computer, such that under this different internal description the
computer is executing a completely different algorithm.  If it is
the abstract computation that matters, then I am suggesting that the
abstract computation being performed is not determined by what happens
in the machine, in the sense that there are completely different
ways of assigning algorithmic descriptions to what happens physically.

I think you are looking at this wrongly.  Instead of thinking of
yourself programming an abstract machine, you should be thinking of
what you are doing as using a specification language to specifify the
causal operations of the machine at a relatively high level (at a
level well above NAND gates).  In fact, most computer programs contain
parts which are machine specific.  There is a cottage industry involved
in porting software from one hardware platform to another.  If we
were really dealing with mathematical idealizations, there should be
no need for such porting.

When I program, I have to concern my self with the bit size of
integers; with the precision of floating point numbers; with the best
way of representing decimal data in this particular machine; with
converting between the machine byte order of data and the network
byte order; with whether files are organized as a stream of bytes or
a sequence of records; with whether file locking protocols apply to a
region of bytes within a file, to physical stored record block, or to
the whole file; with whether output data is immediately transmitted
or is held in a buffer for delayed processing.  None of these things
would matter if I were working with an idealized abstract
computation.

I think you are deceiving yourself.

In <6htajh$...@ux.cs.niu.edu>, rick...@cs.niu.edu (Neil Rickert) writes:
 [snip]

Let's focus on this point very closely.

Consider a very simple computer.   It's called an "OR gate".   It has
two external inputs, and one external output.

The function it computes is "logical OR".   If either of the inputs is
active, the output is active.  If both inputs are inactive, the output
is inactive.

I say that the function it computes is determined by whatever is inside
the black box of the computer, and will remain the same no matter how
you choose to describe it.  The function will remain the same even if
can find no use for it, or if you think it is computing NOT(NOT A AND
NOT B).  It does what it does, regardless.

If you agree, then I'll add more inputs, and complicate the function a
little.  I'll claim that nothing has changed, that the function being
computed is still dependent on what is inside the box, not on your
description.  I'll claim that this holds regardless of how complex we
make that internally-computed function.

What do you say?

Bill

Every time I check in here, I find you guys having the same silly
parochial disagreements about who is being more clumsy or idiosyncratic
(and "a Turing machine computes" is strikingly clumsy -- your professed
ignorance concerning the universe of authors cannot be construed as
*support* for using the phrase; a Turing machine is a *description* of a
computation).

For an *informed* discussion of the issues of computation, cognition,
instantiation, and their relationship, I refer you once again to

http://ling.ucsc.edu/~chalmers/papers/computation.html

(and I will mention, once again for those who confuse people with their
ideas, that this recommendation is not an endorsement of Chalmer's
anti-materialism).

--
<J Q B>

If this is a strong argument, I hate to think what a weak one would be.
This "argument" fails to support the critical claim that conscious
minds are not algorithm-based and fails to show that algorithm-based
methods cannot achieve something achieved in some other way.
It even contains its own refutation: neural nets are commonly simulated.

OTOH, "computer hardware" includes things like photosensors
that perform important functions that their simulations don't.

--
<J Q B>