chess, simulation, and artificial str^h^h^hlife
Pete Dunkelberg
>...
>A computer program 'playing chess' need not use an incomplete or
>simplified model for the simple reason that chess can be precisely
>described in a small number of unambiguous expressions.
Model of what? I think we agree, but I would like to add:
Chess programs do not model human thought, nor do they claim to. The
human expert typically looks from 1 to 5 moves by each side ahead
then uses positional judgment, based on seeing the (future) position as
a whole and liking it or not. No one has come close to 'precisely
describing' positional judgement 'in a small number of unambiguous
expressions'.
A program calculates MANY more moves. Why must it do that? Because it
has no analog for seeing the position as a whole, and there is no
algorythm for judgement. Instead of judging a position as a whole, the
computer makes a mathematical calculation on elements of the future
position. This is so inferior to a master's judgement that the computer
must check orders of magnitude more future variations.
I think this supports your overall point.
Pete
Tim Tyler
: And that is the way in which AL differs from RL: it will always be an
: approximation, because observations will never be complete and so the
: mathematical descriptions will always be partial.
Your arguments about the eternal difference between 'real' and
'artificial' life appear to assume that 'artificial' life exists only in
the form of computer models, or in virtual worlds.
Some researchers are interested in creating 'real' artificial life:
physical self-replicating systems undergoing natural evolutionary
processes in the environment of the real physical world.
I assume that if you were presented with such a species of artificial
life, you would be prepared to grant it equal status with existing
biological life, rather than calling it 'just a model' - or that you view
this scenario as being impossible.
--
__________
|im |yler The Mandala Centre http://www.mandala.co.uk/ t...@cryogen.com
Pain is just God's way of hurting you.
Pete Dunkelberg
>Your arguments about the eternal difference between 'real' and
>'artificial' life appear to assume that 'artificial' life exists only in
>the form of computer models, or in virtual worlds.
>
>Some researchers are interested in creating 'real' artificial life:
>physical self-replicating systems undergoing natural evolutionary
>processes in the environment of the real physical world.
>
non-fiction?? book _The Age of Spiritual Machines_.
Stewart Dean
>>>>>> Pete Dunkelberg pounded silicon into:
>
> > Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
> >> ... A computer program 'playing chess' need not use an incomplete
> >> or simplified model for the simple reason that chess can be
> >> precisely described in a small number of unambiguous expressions.
>
> > Model of what?
>
>In chess simulation, as you describe, the model is of the legal game
>tree.
>
> > I think we agree, but I would like to add: Chess programs do not
> > model human thought, nor do they claim to.
>
>As an author of a (now obsolete) chess program, I agree.
>
> > The human expert typically looks from 1 to 5 moves by each side
> > ahead then uses positional judgment, based on seeing the (future)
> > position as a whole and liking it or not. No one has come close
> > to 'precisely describing' positional judgement 'in a small number
> > of unambiguous expressions'. A program calculates MANY more moves.
>
>Agreed.
>
> > Why must it do that? Because it has no analog for seeing the
> > position as a whole, and there is no algorythm for judgement.
>
>Not exactly true, but the difference is borderline and not relevant to
>your point. Current chess programs use various strategies, we could
>label as 'move elimination heuristics', to reduce the search tree to a
>managable size. These involve a partial embedding of position
>and move 'quality' judgement. However, even the best heuristics still
>only consider a part of a position in making the selection.
>
> > Instead of judging a position as a whole, the computer makes a
> > mathematical calculation on elements of the future position. This
> > is so inferior to a master's judgement that the computer must
> > check orders of magnitude more future variations.
>
>Agreed.
>
> > I think this supports your overall point.
>
>thanks. I think so as well ;-)
>
The player is a simulation - agreed. The game remains 'real' as it
only requires the bits be moved according to the rules. Could be
legally random - it'd still be a game of chess.
Stewart Dean - ste...@webslave.dircon.co.uk
alife guide - http://www.webslave.dircon.co.uk/alife
Stewart Dean
Fact often is stranger than fiction :) (it just lags behind a bit in
the case of Sci Fi)
Cheers
Tim Tyler
:>>>>> Tim Tyler pounded silicon into:
: > Your arguments about the eternal difference between 'real' and
: > 'artificial' life appear to assume that 'artificial' life exists
: > only in the form of computer models, or in virtual worlds.
: No. They don't assume that is the only form of artifical life. They
: are merely arguments about *just* that form of artificial life. [...]
I see - I missed your qualifying paragraph :-/
: > Some researchers are interested in creating 'real' artificial
: > life: physical self-replicating systems undergoing natural
: > evolutionary processes in the environment of the real physical
: > world.
: I am vaguely aware of these [...] do you, by the way, have a good
: pointer to this class of experiment?
Hmm. I view many people who are trying to create artificial life inside
computers as fitting into this caregory.
I doubt that life forms will be satisfied living in a narrow niche which
is in potential danger of following its creators into extinction. If life
gets a decent foothold inside virtual worlds, then it will be almost bound
to want to escape into new niches, as soon as it finds out about the
existence of 'the outside world'.
Artificial self-replicating systems were originally discussed in some
detail by John von-Neumann.
Many people interested in nanotechnology view self-replication as of
fundamental importance to their work:
http://sandbox.xerox.com/nano/ (general nano links)
Robot makers, such as Hans Moravec may consider very large living systems,
composed of robots, computers and possibly, intially man (in symbiosis) to
be prominent candidates for artificial life forms:
http://www.frc.ri.cmu.deu/~hpm/
A.G. Cairns-Smith has suggested that primitive life may be forming
continuously and that the best place to look for it may be in deposits
of clay minerals:
http://www.chem.gla.ac.uk/~bobh/agcs.html (fairly mundane home page).
"Various folks" are also trying to make rather boring artificial DNA based
organisms...
--
__________
|im |yler The Mandala Centre http://www.mandala.co.uk/ t...@cryogen.com
Fatal error: you're dead.
Pete Dunkelberg
>In chess simulation, as you describe, the model is of the legal game
>tree.
>
> > The human expert typically looks from 1 to 5 moves by each side
> > ahead then uses positional judgment, based on seeing the (future)
> > position as a whole and liking it or not. No one has come close
> > to 'precisely describing' positional judgement 'in a small number
> > of unambiguous expressions'. A program calculates MANY more moves.
>
>Agreed.
>
> > Why must it do that? Because it has no analog for seeing the
> > position as a whole, and there is no algorythm for judgement.
>
>Not exactly true, but the difference is borderline and not relevant to
>your point. Current chess programs use various strategies, we could
>label as 'move elimination heuristics', to reduce the search tree to a
>managable size. These involve a partial embedding of position
>and move 'quality' judgement. However, even the best heuristics still
>only consider a part of a position in making the selection.
>
> > Instead of judging a position as a whole, the computer makes a
> > mathematical calculation on elements of the future position. This
> > is so inferior to a master's judgement that the computer must
> > check orders of magnitude more future variations.
>
>Agreed.
>
Glad to hear from someone who knows how the programs work.
I'd be interested in your comments or corrections of the following.
My impression is that in order to eliminate a move, and all it's
consequences from further consideration, the program must first consider
it - in each new position, every move must be considered.
The human player not only just looks a few moves ahead, he only
considers a few 'candidate' moves at each turn. Also remember that a
master plays at nearly master level even in speed chess.
My impression is that the person does the opposite of trimming the tree:
he SPROUTS a tree, but a much smaller one. But the first step is to
apprehend the position as a whole.
Why is the palyer so unlike a computer? Because he EVOLVED.
Neurons are much slower than microprocessors, and earlier organisms had
far fewer of them than we now enjoy. The 'position', i.e. surroundings,
had to be evaluated in real time all the same. Nothing like exhaustive
calculation was an option, yet things that mattered had better be
noticed...
Pete
Richard Harter
This is incorrect or, to be persnickety, it is correct but misleading.
In a typical chess position there may be twenty or thirty legal moves.
A chess program will rate them using programmed heuristic evaluation
formulas. Most of them will be discarded immediately. The program will
grow a move tree which will be dynamically pruned.
>The human player not only just looks a few moves ahead, he only
>considers a few 'candidate' moves at each turn. Also remember that a
>master plays at nearly master level even in speed chess.
>My impression is that the person does the opposite of trimming the tree:
>he SPROUTS a tree, but a much smaller one. But the first step is to
>apprehend the position as a whole.
This is misleading. The real difference is that the human plans. In
other words the human asks: What goals are feasible in this position;
given that, what moves contribute towards those goals.
>Why is the palyer so unlike a computer? Because he EVOLVED.
>Neurons are much slower than microprocessors, and earlier organisms had
>far fewer of them than we now enjoy. The 'position', i.e. surroundings,
>had to be evaluated in real time all the same. Nothing like exhaustive
>calculation was an option, yet things that mattered had better be
>noticed...
You have to remember that a human being has ~100 billion neurons with a
fan-out on the order of 1000 and that it is intrisically wired for
parallel processing. Yes, it evolved for real time operation, but the
brain is one hell of an impressive bit of computer architecture.
The ordinary computer, no matter how fast it is, is handicapped by the
bottleneck of a CPU.
Richard Harter, c...@tiac.net, The Concord Research Institute
URL = http://www.tiac.net/users/cri, phone = 1-978-369-3911
We do not ask of the enlightened one what mountains he has moved.
It suffices that he sits there with a mysterious smile.
Jonathan Stone
|> > You have to remember that a human being has ~100 billion neurons
|> > with a fan-out on the order of 1000 and that it is intrisically
|> > wired for parallel processing. Yes, it evolved for real time
|> > operation, but the brain is one hell of an impressive bit of
|> > computer architecture.
|>
|> very poorly wired for cooperative concurrent programming. This is a
|> small part of what Hillis failed to recognize that led to the
|> inevitable decline of Thinking Machines.
Yo! Keep the faith! stop by the mid-peninsula sometime and i'll get
you a Guinness.
|> > The ordinary computer, no matter how fast it is, is handicapped by
|> > the bottleneck of a CPU.
|>
|> We've changed that. The next generation of systems will be
|> handicapped by the bottleneck of cache/memory architectures. ;-(
Um, no. The *last* generation is handicapped by the memory-system
architecture. If we`re really lucky, the next generation will have the
memory hierarchy (~64M) on-chip. To get more RAM we'll just get more CPUs.
I'll stop now before I start on about my thesis...
Jonathan Stone
|> You have to remember that a human being has ~100 billion neurons with a
|> fan-out on the order of 1000 and that it is intrisically wired for
|> parallel processing. Yes, it evolved for real time operation, but the
|> brain is one hell of an impressive bit of computer architecture.
richard, you're being innumerate. Compare the gigaflops of an Alpha
to the milli-flops of a hunman. Inteeer arithmetic is an even more
dismal story.
face recognition, on the other hand, now you're talkig.
|> The ordinary computer, no matter how fast it is, is handicapped by the
|> bottleneck of a CPU.
No. I tell you three times. No, make that nine.
You have that completely backwards.
Jonathan Stone
[snip lots and lots]
|> The player is a simulation - agreed. The game remains 'real' as it
|> only requires the bits be moved according to the rules. Could be
|> legally random - it'd still be a game of chess.
I'm happy to agree to that.
I still think a record of a game is distinct from an actual game,
though. A pair of players could inadvertently play the *same* game;
that'd be a game. But merely reading it or moving through the record
isnt a game in my book. I'm not sure I can pin down why in a way
that'd make sense to people who think it is a game.
But the record itself isnt a game -- no more than, in my book,
a videotape of the game is an acutal game itself, or a re-enactment
or recording of an actuar event is *really* the event. It *was* a
game, once, but the reproduction isnt the acutal thing.
(thats as clear as I can get, right now.)
Jonathan Stone
|> I know this, and you know this, but my employer refused to believe me
|> when I tried to explain this to them. It is amazing how much wishful
|> thinking you can do when you know that you can't fix the real
|> bottleneck. (Various bright people tried to convince me that the
|> solution was 'bigger caches.')
|>
|> 64M of on-chip memory isn't enough for the next generation. According
|> to my best-guess calculations, assuming a uniprocessor, 256m *might*
|> be enough, but 128m moves the bottleneck from the cpu/memory path to
|> the memory/i-o path.
Oh. Maybe I *should* talk about my thesis.
|> if you think they're wishfully thinking about cpu/memory, look for
|> 'I2O' to see how best to do a *really bad* design of an i/o subsystem:
|> I2O combines the worst of bus-based I/O with the worst of "IOP"
|> design. :-(
I've never stopped being amazed to find hardware engineers who've
never haerd of the ``Wheel of Reincarnation'' from Evans and
Sutherland (from the 60s? Early 70s?). Not to even mention XTP.
Anyway, we should take this offline.
Stew Dean
(Jonathan Stone) wrote:
Clear, understood and agreed. Sounds much like the chinese room
syndrome - or a computer going through a sequential list of
instructions.
Stew Dean
(Jonathan Stone) wrote:
What about parallel processing guys - wasnt that what thinking
machines where all about? Look at the new supercomputers being
created (especially that recent Linux based on costing a fraction of
the atom bomb beasts)
I know it aint easy but a handful of 286s could surely outperfom one
top of the range pentium - why is everyone still trying to build one
amazing machine rather than just using a cellular system
(decentralised - like, er, life)?
Instead of sending one robot to mars and curse when it breaks down why
not send loads of small ones and let em all do bit of the job - doesnt
matter if a few go awol (this is an old idea but so far everybody
still appears to be thinking in a top down way!)
You can have amazing machines based upon the crap buses, small cache
and low clock rates of today - in theory.
Matt Silberstein
jona...@Whisk.DSG.Stanford.EDU (Jonathan Stone):
>In article <36b0ae4a...@news.dircon.co.uk>, ste...@webslave.dircon.co.uk (Stewart Dean ) writes:
>
>[snip lots and lots]
>
>|> The player is a simulation - agreed. The game remains 'real' as it
>|> only requires the bits be moved according to the rules. Could be
>|> legally random - it'd still be a game of chess.
>
>I'm happy to agree to that.
>
>I still think a record of a game is distinct from an actual game,
>though. A pair of players could inadvertently play the *same* game;
>that'd be a game. But merely reading it or moving through the record
>isnt a game in my book. I'm not sure I can pin down why in a way
>that'd make sense to people who think it is a game.
>
I think it has to do with the complexity of the action leading to the
next move. And whether the next move is actually affected by the
current move. I we are moving through the record I only need to look
at the next move on the record. And I would do so even if you made a
move not on the record.
>But the record itself isnt a game -- no more than, in my book,
>a videotape of the game is an acutal game itself, or a re-enactment
>or recording of an actuar event is *really* the event. It *was* a
>game, once, but the reproduction isnt the acutal thing.
>(thats as clear as I can get, right now.)
Has anyone argued that a record of the same was the same as the game?
If so, I missed it. The question is what is a game vs what is a
simulated game. Ironically the record of either would look the same:
simply a list of moves.
Matt Silberstein
-------------------------------------------------------
Perfection (in design) is achieved not when there is nothing
more to add, but rather when there is nothing more to take away.
Antoine de Saint-Exupéry (who was an aviator and aircraft designer
when he wasn't being the author of classic children's books)
Pete Dunkelberg
>Somehow, chess players build a 'chess grammar' that allows them to
>recognize groups of pieces as a single thing, the way we learn to see
>whole words without thinking about the syllables. *but* it only works
>for legal moves.
Interesting. And probably related 'somehow' to the next idea:
>
> > Why is the palyer so unlike a computer? Because he
> > EVOLVED. Neurons are much slower than microprocessors, and earlier
> > organisms had far fewer of them than we now enjoy. The
> > 'position', i.e. surroundings, had to be evaluated in real time
> > all the same. Nothing like exhaustive calculation was an option,
> > yet things that mattered had better be noticed...
>
>there's something to that, although the cog-psyc guys are having a
>hell of time figuring out what.
>
Yes, Indeed. and
Thanks to all for the info on chess programs, etc.
BTW, if you haven't seen this yet, you'd like it:
The COGS gang at Sussex, especially Adrian Thompson.
http://www.cogs.susx.ac.uk/users/adrianth/index.html
If you are not aware of his remarkable "evolution on a
chip", Read It! Rather than simulating evolution in software, he
successively modified a population of chips to evolve the ability
to distinguish two inputs. The result is quite different than you
would get with software using the same evolution algorithm, and
partly skips the binary logic that the chip is built for... he has
given up trying to figure out how it works!
Or if you have read it, does anyone have an update?
Pete
Richard Harter
>In article <36b28f9c...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
>
>|> You have to remember that a human being has ~100 billion neurons with a
>|> fan-out on the order of 1000 and that it is intrisically wired for
>|> parallel processing. Yes, it evolved for real time operation, but the
>|> brain is one hell of an impressive bit of computer architecture.
>
>
>richard, you're being innumerate. Compare the gigaflops of an Alpha
>to the milli-flops of a hunman. Inteeer arithmetic is an even more
>dismal story.
Johnathan, that is a really inane comment (about innumerate). There is
an almost total disconnect between what I said and your comment.
>face recognition, on the other hand, now you're talkig.
>|> The ordinary computer, no matter how fast it is, is handicapped by the
>|> bottleneck of a CPU.
>
>No. I tell you three times. No, make that nine.
>You have that completely backwards.
Tell me as many times as you like. Doesn't make it so.
Jonathan Stone
|> jona...@Whisk.DSG.Stanford.EDU (Jonathan Stone) wrote:
|>
|> >In article <36b28f9c...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
|> >
|> >|> You have to remember that a human being has ~100 billion neurons with a
|> >|> fan-out on the order of 1000 and that it is intrisically wired for
|> >|> parallel processing. Yes, it evolved for real time operation, but the
|> >|> brain is one hell of an impressive bit of computer architecture.
|> >
|> >
|> >richard, you're being innumerate. Compare the gigaflops of an Alpha
|> >to the milli-flops of a hunman. Inteeer arithmetic is an even more
|> >dismal story.
|>
|> Johnathan, that is a really inane comment (about innumerate). There is
|> an almost total disconnect between what I said and your comment.
Richard,
You claimed the brain was one hell of an impressive bit of computer
architecture. Yet its performance as a computer is off by several
orders of magnitude from _any_ of the accepted metrics in the field.
Can you provide full bibliographic references to the peer-reviwed
computer architecture literature showing *computer architecture*
metrics of goodness, by which the human brain is an impressive
*computer* architecture?
Jonathan Stone
|> In talk.origins I read this message from
|> jona...@Whisk.DSG.Stanford.EDU (Jonathan Stone):
|>
|> >In article <36b0ae4a...@news.dircon.co.uk>, ste...@webslave.dircon.co.uk (Stewart Dean ) writes:
|> >
|> >[snip lots and lots]
|> >
|> >|> The player is a simulation - agreed. The game remains 'real' as it
|> >|> only requires the bits be moved according to the rules. Could be
|> >|> legally random - it'd still be a game of chess.
|> >
|> >I'm happy to agree to that.
|> >
|> >I still think a record of a game is distinct from an actual game,
|> >though. A pair of players could inadvertently play the *same* game;
|> >that'd be a game. But merely reading it or moving through the record
|> >isnt a game in my book. I'm not sure I can pin down why in a way
|> >that'd make sense to people who think it is a game.
|> >
|> I think it has to do with the complexity of the action leading to the
|> next move.
But that leads to a very slipperly slope with a very poorly defined
metric. Just what is ``complexity''? how do you measure it?
How is it affected by the computing engine used?
I've asked several times whether a paper-and-pencil computation of a
mathematical model counts as "alive". I still ahvent seen an answer.
I've asked repeatedly whether Go\"del numbers encoding a computation
are alive; i havent seen any answer to that, either.
|> And whether the next move is actually affected by the
|> current move.
i think that's an even more dangerous slippery slope. What does it
*mean* for a move ot be affected by the `current move' Are you
talking bout guessing at the intents of the players and why they
choose to make particular moves? Or are you assuming an oracle which
tells you that? Or that it makes sense to an expert chess-player?
(would randomly-generated strings of valid moves with no
apparent purpose ocunt as a game?)
i really dont intend ot be insulting, but htis strikes me as
a naive set of criteria. I cant see how youd come upw ith a sensible
protocol for applying them.
|> I we are moving through the record I only need to look
[As we? If we?]
|> at the next move on the record. And I would do so even if you made a
|> move not on the record.
|> >But the record itself isnt a game -- no more than, in my book,
|> >a videotape of the game is an acutal game itself, or a re-enactment
|> >or recording of an actuar event is *really* the event. It *was* a
|> >game, once, but the reproduction isnt the acutal thing.
|> >(thats as clear as I can get, right now.)
|>
|> Has anyone argued that a record of the same was the same as the game?
yes, someone did say that.
|> If so, I missed it. The question is what is a game vs what is a
|> simulated game. Ironically the record of either would look the same:
|> simply a list of moves.
Games are a difficult case, in that they're purely about social
conventions. If the game is an abstract one -- the rules are purely
abstract -- then a game with a simulated player taking part is still a
game. Chess, or draught, or blackjack.
but if the game isn't purely formal then, no, a mathematical
simulation isn't the same as the real thing. A mathematical model of
a football game (or a videogame) isnt a *real* football game--
no more than, say, fussball (or Subbuteo table soccer) is a real
game of soccer.
isnt that immediately obvious?
PZ Myers
The brain computes. Therefore, the brain is a kind of computer.
Unfortunately, it is a kind of computer for which the "accepted
metrics" are inadequate. The only reason I can think you might
be arguing with Richard on this matter is that you have some
extremely restrictive definition of what a computer ought to be
able to do, and think our brains are failures because they don't
meet some very narrow set of criteria.
I betcha I can come up with more metrics in which human brains are
better than computers than you can metrics in which computers are
better!
--
PZ Myers
Jonathan Stone
[snip lots]
There's a fair overlap between my own research and what marty says.
Th research group I'm in has several peploe who did research building
parallel machines and spent manyears optimizing *one* small program
from the SPLASH benchmark suite (mp3d) to run well on parallel
processors. (Several of the present and past members have published
in the computer architecture community.)
I concur with what Marty says. Its a hard problem.
|> > You can have amazing machines based upon the crap buses, small
|> > cache and low clock rates of today - in theory.
|>
|> No, actually, you can't. If you don't fix the communication
|> bottlenecks, the CPUS sit there and wait a lot of the time. When you
|> do fix the hardware bottlenecks you end up with software communication
|> bottlenecks. When you finally balance the system, you discover that
|> most of your problems don't have the inherent parallism to take
|> advantage of it.
Just to put some numbers in here: these days, a good FPU can give you
a result a nanosecond, if the memory system can keep up. But
communicating across a network (ala beowulf, or SHRIMP, or U-net, or
whatever) means fielding an interrupt in order to shift from
doing number-crunching ot looking at the newly-arrived packet.
That's on the order of five microseconds. Once the OS or software
gets involved to route the packet to the correct application,
you're talking milliseconds. Thats six orders of magnitude.
So you really want problems that not just parallelize well, but that
*partition* well: so that one CPU just doesnt need to talk to another
one very much. that dont have much communication per gigaflop.
For many domains, its just not in the nature of physics for problems
to partition well: you eventually need to push the local state to the
"neigbouring"' nodes, and then youre smack-bang head-on into
communications costs. (which is one reason I jumped so hard on Richard
Harter. wetware doesn't have much in common with computer
architecture).
Far enough off-topic already?
Pete Dunkelberg
>What is not clear and has never been shown is that the brain *only*
>computes. Roger Penrose argues in two books that it may well be that
>there exist physical phenomena which can not be modeled as computable
>functions that are present in the brain and account for
>consciousness. If Penrose is right, then the brain is literaly 'not
>computable', which means that (at least some of) its functions can not
>be modeled with a computer.
>
I'm not sold on Penrose, but the obvious noncomputable feature is
consciousness itself. As the brain is finite, I am not sure how non
computable functions come into it.
Pete
Richard Harter
>In article <36b37fbd....@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
>|> jona...@Whisk.DSG.Stanford.EDU (Jonathan Stone) wrote:
>|>
>|> >In article <36b28f9c...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
>|> >
>|> >|> You have to remember that a human being has ~100 billion neurons with a
>|> >|> fan-out on the order of 1000 and that it is intrisically wired for
>|> >|> parallel processing. Yes, it evolved for real time operation, but the
>|> >|> brain is one hell of an impressive bit of computer architecture.
>|> >
>|> >
>|> >richard, you're being innumerate. Compare the gigaflops of an Alpha
>|> >to the milli-flops of a hunman. Inteeer arithmetic is an even more
>|> >dismal story.
>|>
>|> Johnathan, that is a really inane comment (about innumerate). There is
>|> an almost total disconnect between what I said and your comment.
>
>Richard,
>
>You claimed the brain was one hell of an impressive bit of computer
>architecture. Yet its performance as a computer is off by several
>orders of magnitude from _any_ of the accepted metrics in the field.
And yet, oddly enough, it handles with ease tasks that no non-organic
computer is able to handle..
>Can you provide full bibliographic references to the peer-reviwed
>computer architecture literature showing *computer architecture*
>metrics of goodness, by which the human brain is an impressive
>*computer* architecture?
I dunno, can you cite any non-organic computer with that many active
computational elements within three orders of magnitude?
Richard Harter
>>>>>> Richard Harter pounded silicon into:
>
>[snip]
>
> > I dunno, can you cite any non-organic computer with that many
> > active computational elements within three orders of magnitude?
>
>how many neurons does it take to make a computational element?
>
>You can't model a turing machine with just one, after all.
A neuron is a computational element. Exactly what a computational
element consists of is deliberately left vague but it does have more
complexity than a nand gate.
Matt Silberstein
<usene...@usenet.nospam.fogey.com>:
>>>>>> PZ Myers pounded silicon into:
>
>[snip]
>
> > The brain computes. Therefore, the brain is a kind of
> > computer. Unfortunately, it is a kind of computer for which the
> > "accepted metrics" are inadequate. The only reason I can think you
> > might be arguing with Richard on this matter is that you have some
> > extremely restrictive definition of what a computer ought to be
> > able to do, and think our brains are failures because they don't
> > meet some very narrow set of criteria.
>
>In mathematics, and computer science, there is a very precise
>definition of 'compute', which is described by Alan Turing's infamous
>"Turing Machine" but which has a very formal basis.
>
And, since we have used words with very precise definitions in those
domains we are, I guess, strictly held to those definitions.
>A digital computer is an aproximation of a Turing Machine, from the
>poitn of view of mathematics, and so, 'computes'.
>
>In that sense, anything that can be modeled mathematically by what is
>known as a 'computable function' can be modeled, at least in theory,
>on a computer.
>
>Computers, on the other hand can, formally, do no more than this.
>
>What is not clear and has never been shown is that the brain *only*
>computes. Roger Penrose argues in two books that it may well be that
>there exist physical phenomena which can not be modeled as computable
>functions that are present in the brain and account for
>consciousness. If Penrose is right, then the brain is literaly 'not
>computable', which means that (at least some of) its functions can not
>be modeled with a computer.
>
You gave, IIRC, a reference to one of Penrose's books. Could you
supply both? BTW, in general, what do you think of his argument?
[snip]
Matt Silberstein
>In article <36b923fd...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
>|> In talk.origins I read this message from
>|> jona...@Whisk.DSG.Stanford.EDU (Jonathan Stone):
>|>
>|> >In article <36b0ae4a...@news.dircon.co.uk>, ste...@webslave.dircon.co.uk (Stewart Dean ) writes:
>|> >
>|> >[snip lots and lots]
>|> >
>|> >|> The player is a simulation - agreed. The game remains 'real' as it
>|> >|> only requires the bits be moved according to the rules. Could be
>|> >|> legally random - it'd still be a game of chess.
>|> >
>|> >I'm happy to agree to that.
>|> >
>|> >I still think a record of a game is distinct from an actual game,
>|> >though. A pair of players could inadvertently play the *same* game;
>|> >that'd be a game. But merely reading it or moving through the record
>|> >isnt a game in my book. I'm not sure I can pin down why in a way
>|> >that'd make sense to people who think it is a game.
>|> >
>|> I think it has to do with the complexity of the action leading to the
>|> next move.
>
>But that leads to a very slipperly slope with a very poorly defined
>metric. Just what is ``complexity''? how do you measure it?
>How is it affected by the computing engine used?
>
>I've asked several times whether a paper-and-pencil computation of a
>mathematical model counts as "alive". I still ahvent seen an answer.
>I've asked repeatedly whether Go\"del numbers encoding a computation
>are alive; i havent seen any answer to that, either.
>
I responded to the second. If you have not seen it, then look. I am
not receiving many posts in this thread, you may have the same
problem. In response to the first, no it is not the same. I am not
sure I can pin down why in a way that'd make sense to people who
disagree.
>
>|> And whether the next move is actually affected by the
>|> current move.
>
>i think that's an even more dangerous slippery slope. What does it
>*mean* for a move ot be affected by the `current move' Are you
>talking bout guessing at the intents of the players and why they
>choose to make particular moves? Or are you assuming an oracle which
>tells you that? Or that it makes sense to an expert chess-player?
>(would randomly-generated strings of valid moves with no
>apparent purpose ocunt as a game?)
>
Then why not give some answers that you think are reasonable.
>i really dont intend ot be insulting, but htis strikes me as
>a naive set of criteria. I cant see how youd come upw ith a sensible
>protocol for applying them.
>
I tried to give a surface answer because it was going off topic. If
you want to go in depth there, fine. BTW, since you find this naive I
guess I can assume you are being naive any time you give a less than
comprehensive response.
>|> I we are moving through the record I only need to look
>[As we? If we?]
>
>|> at the next move on the record. And I would do so even if you made a
>|> move not on the record.
>
>
>|> >But the record itself isnt a game -- no more than, in my book,
>|> >a videotape of the game is an acutal game itself, or a re-enactment
>|> >or recording of an actuar event is *really* the event. It *was* a
>|> >game, once, but the reproduction isnt the acutal thing.
>|> >(thats as clear as I can get, right now.)
>|>
>|> Has anyone argued that a record of the same was the same as the game?
>
>yes, someone did say that.
>
>|> If so, I missed it. The question is what is a game vs what is a
>|> simulated game. Ironically the record of either would look the same:
>|> simply a list of moves.
>
>Games are a difficult case, in that they're purely about social
>conventions. If the game is an abstract one -- the rules are purely
>abstract -- then a game with a simulated player taking part is still a
>game. Chess, or draught, or blackjack.
>
>but if the game isn't purely formal then, no, a mathematical
>simulation isn't the same as the real thing. A mathematical model of
>a football game (or a videogame) isnt a *real* football game--
>no more than, say, fussball (or Subbuteo table soccer) is a real
>game of soccer.
>
Yes, a simulated football game is not a real football came (for most
meanings of football). But no one, AFAICT, has claimed otherwise. It
was the other games, the ones you now dismiss as formal, that were
under discussion.
>isnt that immediately obvious?
Yes, and irrelevant.
Tim Tyler
: In article <36b37fbd....@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
: |> jona...@Whisk.DSG.Stanford.EDU (Jonathan Stone) wrote:
: |> >In article <36b28f9c...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
: |> >|> You have to remember that a human being has ~100 billion neurons with a
: |> >|> fan-out on the order of 1000 and that it is intrisically wired for
: |> >|> parallel processing. Yes, it evolved for real time operation, but the
: |> >|> brain is one hell of an impressive bit of computer architecture.
: |> >
: |> >richard, you're being innumerate. Compare the gigaflops of an Alpha
: |> >to the milli-flops of a hunman. Inteeer arithmetic is an even more
: |> >dismal story.
: |>
: |> Johnathan, that is a really inane comment (about innumerate). There is
: |> an almost total disconnect between what I said and your comment.
: You claimed the brain was one hell of an impressive bit of computer
: architecture. Yet its performance as a computer is off by several
: orders of magnitude from _any_ of the accepted metrics in the field.
Hrumph ;-)
: Can you provide full bibliographic references to the peer-reviwed
: computer architecture literature showing *computer architecture*
: metrics of goodness, by which the human brain is an impressive
: *computer* architecture?
See, for one example, the *advocate* of machine intelligence Hans
Moravec's recent 'Robot' book. By his metrics, desktop computers are
currently somewhere around the cockroach level in terms of processing
power.
http://www.frc.ri.cmu.edu/~hpm/book98/ has the graphs, IIRC.
Human brains massively out-perform modern computers in a wide range of
real-world tasks. Their superior, massively parallel connectionist
architecture probably helps them to do this. Why do you think the
neural net people spend so much time trying to copy nature's work?
--
__________
|im |yler The Mandala Centre http://www.mandala.co.uk/ t...@cryogen.com
May all your hang-ups be drip-dry.
Jonathan Stone
|> Jonathan Stone <jona...@Cup.DSG.Stanford.EDU> wrote:
|> : You claimed the brain was one hell of an impressive bit of computer
|> : architecture. Yet its performance as a computer is off by several
|> : orders of magnitude from _any_ of the accepted metrics in the field.
|>
|> Hrumph ;-)
Sorry? Do you disagree? i dont understand this at all?
|> : Can you provide full bibliographic references to the peer-reviwed
|> : computer architecture literature showing *computer architecture*
|> : metrics of goodness, by which the human brain is an impressive
|> : *computer* architecture?
|>
|> See, for one example, the *advocate* of machine intelligence Hans
|> Moravec's recent 'Robot' book. By his metrics, desktop computers are
|> currently somewhere around the cockroach level in terms of processing
|> power.
|> http://www.frc.ri.cmu.edu/~hpm/book98/ has the graphs, IIRC.
But that is not part of the computer architecture literature. Just
how fast can a cockroach add? How does it do on the SPLASH
benchmarks? Or TPC-C? What's it's Specrate, and how does it scale
across multiple cockroaches? Where can i find peer-reviewed articles
(reviewed by _computer architects_) which compare it to other computer
architectures? Do you have any real cognizance of what computer
architecture is?
Or, perhaps, are you merely speaking metaphorically?
|> Human brains massively out-perform modern computers in a wide range of
|> real-world tasks.
Sure. Nobody ever denied this. That does not make them a computer
architecture, though. One might look carefully at the recent
discussions here about ``design'' and how the word, necessarily,
implies a designer. So, surely, does `architecture'. No?
|> Their superior, massively parallel connectionist
|> architecture probably helps them to do this.
Perhaps they do. That doesnt necessarily make it a computer
architecture; let alone a particularly *good* one.
|> Why do you think the
|> neural net people spend so much time trying to copy nature's work?
No, quite honestly, I dont spend much time wondering why neural net
researchers do anything. Why should i? In copmuter architecture, the
quantitative tradeoffs of computational costs versus communication
costs are orders of magnitude different than with wetware. Thus,
"neural nets" are not especially relevant to computer architecture.
They just arent the same field.
Please re-read what Marty Fouts said about connection machines, and
what I posted about communications costs versus computation costs.
Jonathan Stone
In article <36b8a2ee...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
[snip]
|> >|> >
|> >|> I think it has to do with the complexity of the action leading to the
|> >|> next move.
|> >
|> >But that leads to a very slipperly slope with a very poorly defined
|> >metric. Just what is ``complexity''? how do you measure it?
|> >How is it affected by the computing engine used?
|> >
|> >I've asked several times whether a paper-and-pencil computation of a
|> >mathematical model counts as "alive". I still ahvent seen an answer.
|> >I've asked repeatedly whether Go\"del numbers encoding a computation
|> >are alive; i havent seen any answer to that, either.
|> >
|> I responded to the second. If you have not seen it, then look.
No, I havent seen it.
I note, again, that you havent chdo
|> I am
|> not receiving many posts in this thread, you may have the same
|> problem.
Perhaps.
In response to the first, no it is not the same. I am not
|> sure I can pin down why in a way that'd make sense to people who
|> disagree.
But the two computations *are* formally equivalent. As computations
go, they are the same computatoin. By definition.
If you think they're different, then you are *profoundly* ignorant.
The only real difference is time. If you want to decide that life
depends on the rate at which creatures "live", then (again) you are
profoundly ignorant, since even on earth, metabolic rates vary with
temperature. And besides, your argument was based on _complexity_,
not rates of state change associated with that complexity.
You've been a thorough ignoramus in this whole arugment, Matt.
Jonathan Stone
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
matt,
You really dont understand, do you?
|> >i really dont intend ot be insulting, but htis strikes me as
|> >a naive set of criteria. I cant see how youd come upw ith a sensible
|> >protocol for applying them.
|> >
|> I tried to give a surface answer because it was going off topic. If
|> you want to go in depth there, fine. BTW, since you find this naive I
|> guess I can assume you are being naive any time you give a less than
|> comprehensive response.
No, it's not going off-topic. And the reason I think your answers
are naive is this:
1. you're repeatedly making fundamental, basic mistakes
2. I ask questions or make comments that highlight
those mistakes
3. you dont answer them, or you dont get it, or (in some cases)
the responses get lost.
4. Even so, you go on making the same fundamental mistakes.
Please dont stop. I've been showing some of your gaffes to friends
who visit. They've been laughing hysterically. One visitor (who has
a recent PhD in bio-informatics) told me that your errors are on the
same level as suggesting a polypeptide sequence of heme.
In short, you're making a complete and utter idiot of yourself.
Do you really want to keep on doing that?
Matt Silberstein
jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>[posted, d-and hopefully, emailed]
>
>
>In article <36b8a2ee...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
Jonathan, is someone forging posts with your name on them? Anyway,
while there is some of content here, the bulk of this post is simply
personal comments and I see no reason to try to pick out the little
bit of content and respond. If anyone else wishes to argue these
points, please repeat them. If Jonathan wishes to repost with just the
content, I will respond.
Matt Silberstein
jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>In article <36b8a2ee...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
>|> In talk.origins I read this message from
>|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>
>matt,
>
>You really dont understand, do you?
>
>
>|> >i really dont intend ot be insulting, but htis strikes me as
>|> >a naive set of criteria. I cant see how youd come upw ith a sensible
>|> >protocol for applying them.
>|> >
>|> I tried to give a surface answer because it was going off topic. If
>|> you want to go in depth there, fine. BTW, since you find this naive I
>|> guess I can assume you are being naive any time you give a less than
>|> comprehensive response.
>
>No, it's not going off-topic. And the reason I think your answers
>are naive is this:
>
> 1. you're repeatedly making fundamental, basic mistakes
> 2. I ask questions or make comments that highlight
> those mistakes
> 3. you dont answer them, or you dont get it, or (in some cases)
> the responses get lost.
> 4. Even so, you go on making the same fundamental mistakes.
>
>
>Please dont stop. I've been showing some of your gaffes to friends
>who visit. They've been laughing hysterically. One visitor (who has
>a recent PhD in bio-informatics) told me that your errors are on the
>same level as suggesting a polypeptide sequence of heme.
>
>In short, you're making a complete and utter idiot of yourself.
>
>Do you really want to keep on doing that?
Jonathan, I am glad you have these anonymous authorities on your side.
Matt Silberstein
jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
[snip]
>Sure. Nobody ever denied this. That does not make them a computer
>architecture, though. One might look carefully at the recent
>discussions here about ``design'' and how the word, necessarily,
>implies a designer. So, surely, does `architecture'. No?
>
On-line Webster's.
Main Entry: ar·chi·tec·ture
Pronunciation: 'är-k&-"tek-ch&r
Function: noun
Date: 1555
1 : the art or science of building; specifically : the art or practice
of designing and building structures and especially habitable ones
2 a : formation or construction as or as if as the result of conscious
act <the architecture of the garden> b : a unifying or coherent form
or structure <the novel lacks architecture>
3 : architectural product or work
4 : a method or style of building
5 : the manner in which the components of a computer or computer
system are organized and integrated
By definitions 1 or 4 you are using architecture metaphorically. By
definitions 2,3, and 5 it does not necessarily imply an architect.
I think you have confused "there is a design" with "it was designed".
>
[snip]
Jonathan Stone
|> In talk.origins I read this message from
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|> >But the two computations *are* formally equivalent. As computations
|> >go, they are the same computatoin. By definition.
|> >If you think they're different, then you are *profoundly* ignorant.
|> >
|> >The only real difference is time. If you want to decide that life
|> >depends on the rate at which creatures "live", then (again) you are
|> >profoundly ignorant, since even on earth, metabolic rates vary with
|> >temperature. And besides, your argument was based on _complexity_,
|> >not rates of state change associated with that complexity.
|> >
|> >
|> >You've been a thorough ignoramus in this whole arugment, Matt.
|>
|> Jonathan, is someone forging posts with your name on them?
No.
|> Anyway,
|> while there is some of content here, the bulk of this post is simply
|> personal comments and I see no reason to try to pick out the little
|> bit of content and respond. If anyone else wishes to argue these
|> points, please repeat them.
Nonsense. the absic content is that you dont know what you're talking
about. Your comments make are a laughing-stock.
|>Jonathan wishes to repost with just the
|> content, I will respond.
Matt, you just dont get it. My basic point is that you are profoundly
ignorant here: you simply dont know what you're talking about.
My second point is that you are so ignorant you dont even appreciate
just how foolish your comments are.
My third point is that i'm fed up with *you* asking *me* to correct
your mistakes. You make claims. The onus is on you to make them
coherent and non-self-contradictory. But you don't.
Jonathan Stone
|> In talk.origins I read this message from
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|>
|> >|> In talk.origins I read this message from
|> >|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|> >
|> >
|> >You really dont understand, do you?
|> >
|> >
|> >|> >i really dont intend ot be insulting, but htis strikes me as
|> >|> >a naive set of criteria. I cant see how youd come upw ith a sensible
|> >|> >protocol for applying them.
|> >|> >
|> >|> I tried to give a surface answer because it was going off topic. If
|> >|> you want to go in depth there, fine. BTW, since you find this naive I
|> >|> guess I can assume you are being naive any time you give a less than
|> >|> comprehensive response.
|> >
|> >No, it's not going off-topic. And the reason I think your answers
|> >are naive is this:
|> >
|> > 1. you're repeatedly making fundamental, basic mistakes
|> > 2. I ask questions or make comments that highlight
|> > those mistakes
|> > 3. you dont answer them, or you dont get it, or (in some cases)
|> > the responses get lost.
|> > 4. Even so, you go on making the same fundamental mistakes.
|> >
|> >
|> >Please dont stop. I've been showing some of your gaffes to friends
|> >who visit. They've been laughing hysterically. One visitor (who has
|> >a recent PhD in bio-informatics) told me that your errors are on the
|> >same level as suggesting a polypeptide sequence of heme.
|> >
|> >In short, you're making a complete and utter idiot of yourself.
|> >
|> >Do you really want to keep on doing that?
|>
|> Jonathan, I am glad you have these anonymous authorities on your side.
Matt, you are a hypocrite. You have appealed to the authority of
un-named personal acquaintances yourself. But If you want the names
of my friends, or their relevaent qualifications, you'll get them.
Provided my friends agree. (at least one of them woudlnt want to give
their details to someone who says the stupid things you've been saying.)
Not that that makes what you've been saying any less stupid.
Your statements stand on their own grounds.
Jonathan Stone
|> In talk.origins I read this message from
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|>
|>
|> >Sure. Nobody ever denied this. That does not make them a computer
|> >architecture, though. One might look carefully at the recent
|> >discussions here about ``design'' and how the word, necessarily,
|> >implies a designer. So, surely, does `architecture'. No?
|> >
|> On-line Webster's.
|>
|>
|> Main Entry: ar·chi·tec·ture
|> Pronunciation: 'är-k&-"tek-ch&r
|> Function: noun
|> Date: 1555
|> 1 : the art or science of building; specifically : the art or practice
|> of designing and building structures and especially habitable ones
|> 2 a : formation or construction as or as if as the result of conscious
|> act <the architecture of the garden> b : a unifying or coherent form
|> or structure <the novel lacks architecture>
|> 3 : architectural product or work
|> 4 : a method or style of building
|> 5 : the manner in which the components of a computer or computer
|> system are organized and integrated
|>
|> By definitions 1 or 4 you are using architecture metaphorically.
We have been consistently discussing computer architecture. Meaning
#5. Which, etymologically, derives from meaning #1.
Both of these imply design.
|> By
|> definitions 2,3, and 5 it does not necessarily imply an architect.
No. Wrong. 1 and 5 imply an architect. The only one which explicitly
doesn't require a designer is #2, with the `as if'. #3, #4, and #5
are related to #1; i see them as also implying a designer, tho' I'd
prefer to seew hat Big Oxford has to say. (Websters is not a very good
dicitionary for such purposes).
|> I think you have confused "there is a design" with "it was designed".
I think you are confused on many points, including this one.
Richard Harter
>>>>>> Richard Harter pounded silicon into:
>
> > Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
>
> >>>>>>> Richard Harter pounded silicon into:
> >> [snip]
> >>
> >> > I dunno, can you cite any non-organic computer with that many
> >> active computational elements within three orders of magnitude?
> >>
> >> how many neurons does it take to make a computational element?
> >>
> >> You can't model a turing machine with just one, after all.
>
> > A neuron is a computational element. Exactly what a computational
> > element consists of is deliberately left vague but it does have
> > more complexity than a nand gate.
>
>This may explain some of the differences here. To a computer
>scientist or mathematician, a 'computational element' is a device of
>sophisticated complexity to be able to model a Turing machine, and
>therefore able to "compute."
Nonsense.
>Unless we have a metric which is independent of the implementation by
>which we can determine what you have left vague, there's no way to do
>an apples-to-apples comparison.
What I really had in mind was that a neuron has more functionality than
a simple logic element, e.g. a nand gate. How much more is distinctly
open to question. If a neuron can be modeled by 10 logic elements (and
it may require more) then a hardware equivalent of the brain would
require on the order of one trillion circuit elements.
Matt Silberstein
jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>In article <36b9002e...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
[snip]
>
>|> I think you have confused "there is a design" with "it was designed".
>
>I think you are confused on many points, including this one.
As good an example of an ad hominem fallacy as one could hope for. One
of the above statements comments on the content of a post, the other
on the poster.
Jonathan Stone
|> Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
|> >This may explain some of the differences here. To a computer
|> >scientist or mathematician, a 'computational element' is a device of
|> >sophisticated complexity to be able to model a Turing machine, and
|> >therefore able to "compute."
|>
|> Nonsense.
|> >Unless we have a metric which is independent of the implementation by
|> >which we can determine what you have left vague, there's no way to do
|> >an apples-to-apples comparison.
|>
|> What I really had in mind was that a neuron has more functionality than
|> a simple logic element, e.g. a nand gate. How much more is distinctly
|> open to question. If a neuron can be modeled by 10 logic elements (and
|> it may require more) then a hardware equivalent of the brain would
|> require on the order of one trillion circuit elements.
Right. And a trillion circuit elements is on the order of 100 Gbytes.
In other words, one order of magnitude more than a large database
server, and only two orders of magnitude more than a well-configured
desktop. (yes, I know people with gigabyte machines on their desktop.)
Yet the human brain does not come within eight orders of magnitude of
the computing ability of such boxes. So much for brains being a good
parallel computer architecture.
And we dont know how to compare human tasks like face recognition,
because we still dont know *HOW* human brains recognise faces, so we
cannot model it. Thus, we jut cannot do it efficiently on computers,
because we dont know *how*. Thus, conclusions about computer
architectures for emulating certain human tasks are premature.
Just look at chess: computer chesss programs certainly dont do it the
way humans do, yet research into game-tree pruning coupled with ever
exponentially-faster hardware means computers now outcompete even
the best humans.
Like i said, computer architecture isnt the same as how brains work.
Jonathan Stone
|> It is my conjecture that the neural-net community, and much of the
|> rest of the AI community are failing to make progress because they are
|> up against the computational equivalent of squaring the circle. The
|> answer hinges, as Roger Penrose has pointed out, on whether or not all
|> physical processess can be modeled by what mathematicans call
|> 'computable functions' in the sense of Turing. If not, and the brain
|> relies on such non-computable functions, then neural-nets are a dead
|> end.
This is an interesting conjecture. I dont know of any actual positive
evidence for it, though. Certainly there are philosophers of mind who
are very very skeptical of Penrose' work.
Refutations of the Church-Turing thesis will be eagerly awaited.
|> Now, before someone raises the objection that it should be obvious
|> what is computable and what isn't, let me point out that it took over
|> 2000 years for the necessary mathematics to arise to be able to
|> formally prove that one cannot square the circle.
Sure. Thats very very different from (for example) saying that
computation is alive, due to its complexity and a homomorphism to
biological organisms yet that another instance of the identical
computation isn't alive.
Jonathan Stone
|> What I really had in mind was that a neuron has more functionality than
|> a simple logic element, e.g. a nand gate. How much more is distinctly
|> open to question. If a neuron can be modeled by 10 logic elements (and
|> it may require more) then a hardware equivalent of the brain would
|> require on the order of one trillion circuit elements.
What do we mean by neuron? Real neurons in real brains, or abstract,
simulated `neurons' in an ANN?
Modulo the fact that one is analog (continuous) and the other is
digital, I see little to choose from between an ANN neuron and a gate;
see Minsky's famous perceptron result. so if you want to compare
them, we're down to couting inputs and comparators and stuff.
Nasty.
But all that ignores quite fundamental constraints on VLSI digital
computer design. It's very easy and very cheap to produce large arrays
of identical components. Crossing wires, on the other hand, is very
expensive.
Connectionist architectures assume crossing wires (and interconnects
in general) are cheap. Nobody has yet come up with good ways to build
such hardware.
I just dont think they're commensurate.
Jonathan Stone
|> In talk.origins I read this message from
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|>
|> >
|> >I think you are confused on many points, including this one.
|>
|> As good an example of an ad hominem fallacy as one could hope for. One
|> of the above statements comments on the content of a post, the other
|> on the poster.
No, matt. Its a statement of fact. You *are* confused about many
points. I explained *exactly* why I disagreed with you.
Yet you snippped that out, and you havent responded to my reasons.
Or do you think that disagreeing with you is an ad-hominem?
Jonathan Stone
This is getting ridicuolus. Here is the list of questions again,
for anyone who hasn't noticed that Matt is consistently avoiding
answering them:
1. Matt's definitions of `life' are metaphorical.
THe interpretation of `metabolize' and `reproduce' is so loose
that, as i see it, it includes flames and crystals.
Are flames and crystals alive?
2. Living things do not have on/off switches or reset buttons.
They cannot be paused (e.g, by powerfail or checkpoint
mechanisms) and then restarted.
Computer simluations can. How, then, are computer simulations
alive?
3. Computer simulations do not acutally metabolize.
The level of abstraction wehre we find similarties between
a living thing and a computer simluation of a living
thing does not actually exchange energy: instead, it
exchanges information. (Any energy exchagne takes place
purely within the computational devices themselves).
Precisely how does exchange of information constitue
*real* `metabolism' as opposed to a *smulation* of metabolism?
4. Computer simluations are merely comptuational models.
They can be executed on a huge range of computational devices,
all the way from supercomputers to pencil and paper.
Mathematically, these computations are all identical:
that is the Church-Turing thesis.
Is a computation done on pencil and paper alive?
If not, then Matt's orignal claim is refuted; there
is something to life *APART* from the complexity and
a homomorphism between biological organisms and simluations
of biological organisms.
5. Which goedel numbers of computations are alive, and which
are not, and why?
6. Computer simulations are, by definition, mathematical models
of the things being simulated. The models abstract away
from the real thing. They deal only with some mathematical
aspects of the thing being modelled.
They are not the thing itself.
For example, a simulation of a nuclear power-plant
may model the nuclear reactions and the energy transfer,
but the simulated energy cannot power a real light-bulb.
Nor can a simulation of an aircraft acutally tranport people
or things. ``the map is not the territory''.
Matt's basic thesis is that, in the case of life,
the map *is* the territory. This seems to me, and to Marty,
to be a fundamental semantic error.
What is so special about life that, unliek all other
simluations, the map really *is* the territory?
I've asked Matt these questions repeatedly. He has not answered any
except 5 (and I still havent seen that). His only response is to ask
*me* to answer the questions. Even though I've repeatedly, that I
find that exceedingly offensive (because, as you can guess from the
questions, I think Matt's idea is bloody stupid, and I dont *want* to
try and fix it up.)
It's really up to Matt. He can either try and answer the questions
substantively, or not.
We're all adult enough to see who is being intellecutally dishonest
here, and who isn't.
Jonathan Stone
|> If not, and the brain
|> relies on such non-computable functions, then neural-nets are a dead
|> end.
more than that: any Turing-complete approach would be a dead end.
|> Now, before someone raises the objection that it should be obvious
|> what is computable and what isn't, let me point out that it took over
|> 2000 years for the necessary mathematics to arise to be able to
|> formally prove that one cannot square the circle.
This'd be a lot more plausible if someone could come up with a
function that was not recursively enumerable, and with a model for
``computation'' which could compute it.
Its kinda suggestive that so many formalisms turn out to be equivalent:
* Turing machines
* lambda-calculus
* recursive functions
* Chomsky type-0 languages
* Post production systems
and of course one can prove Go\"dels theorem as a semi-trivial
consequence of the halting problem. so perhaps we should throw in set
theory and arithmetic as well.
Did I miss any of the classic ones?
Richard Harter
>>>>>> Richard Harter pounded silicon into:
>
> > Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
>
> >>>>>>> Richard Harter pounded silicon into:
> >> Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
> >>
> >> >>>>>>> Richard Harter pounded silicon into: [snip] I dunno, can
> >> you cite any non-organic computer with that many active
> >> computational elements within three orders of magnitude? how
> >> many neurons does it take to make a computational element? You
> >> can't model a turing machine with just one, after all.
> >>
> >> > A neuron is a computational element. Exactly what a
> >> computational element consists of is deliberately left vague but
> >> it does have more complexity than a nand gate.
> >>
> >> This may explain some of the differences here. To a computer
> >> scientist or mathematician, a 'computational element' is a device
> >> of sophisticated complexity to be able to model a Turing machine,
> >> and therefore able to "compute."
>
> > Nonsense.
>
>Well that was a useful remark. Would you care to do more than dismiss
>the entire mathematical basis of computation with a single word?
The nonsense lies in saying that to a computer scientist or a
mathematician a computational element *is* a device et cetera. "A
computational element" isn't a standard term in formal treatments. I
don't doubt that someone uses the way you did - people are always
inventing neologisms. None-the-less. You follow this with an
egregious non sequitor, to wit your interpretation of "nonsense" as
"dismiss the entire mathematical basis of computation".
> >> Unless we have a metric which is independent of the
> >> implementation by which we can determine what you have left
> >> vague, there's no way to do an apples-to-apples comparison.
>
> > What I really had in mind was that a neuron has more functionality
> > than a simple logic element, e.g. a nand gate. How much more is
> > distinctly open to question. If a neuron can be modeled by 10
> > logic elements (and it may require more) then a hardware
> > equivalent of the brain would require on the order of one trillion
> > circuit elements.
>
>The estimate that human brains contain three orders of magnitude more
>of these 'logic circuit' equivalents is off by quiet a bit, then. The
>largest extent computer systems have > 1000 nodes where each node has
>> 100,000,000 'logic elements' which would put it on a par with a
>human brain.
I'm not sure what you are counting here. What constitutes a node and
what kind of systems are you talking about? What are you counting as a
logic element? I will cheerfully concede that if you network enough
machines together you can accumulate enough components to match the
neuron count of a single human brain. It remains, though, that a single
machine has orders of magnitude fewer components than the brain. (The
difficulties of comparing components are also cheerfully conceded.)
Richard Harter
>>>>>> Jonathan Stone pounded silicon into:
>
> > In article <wkn22yd...@usenet.nospam.fogey.com>, Marty Fouts
> > <usene...@usenet.nospam.fogey.com> writes:
>
> > |> If not, and the brain |> relies on such non-computable
> > functions, then neural-nets are a dead |> end.
>
> > more than that: any Turing-complete approach would be a dead end.
>
>Yup.
>
> > |> Now, before someone raises the objection that it should be
> > obvious |> what is computable and what isn't, let me point out
> > that it took over |> 2000 years for the necessary mathematics to
> > arise to be able to |> formally prove that one cannot square the
> > circle.
>
> > This'd be a lot more plausible if someone could come up with a
> > function that was not recursively enumerable, and with a model for
> > ``computation'' which could compute it.
>
>Huh? The point was that it took 2000 years to come up with the
>mathematical system necessary to show that the circle couldn't be
>squared, so that it is likely to take a while to come up with the
>function that is not recursively enumerable.
It's sort of an apples and oranges comparison. If Church's thesis is
correct (and the nifty thing is that if it is you can't prove it) there
ain't no such animal. Saying that we can find a way to compute
functions which are not recursively enumerable is like saying that we
can find a way to travel faster than light. Yeah, maybe, but that's not
the way to bet.
> > Its kinda suggestive that so many formalisms turn out to be
> > equivalent: Turing machines lambda-calculus recursive functions
> > Chomsky type-0 languages Post production systems
>
>Suggestive of what, though?
>
> > and of course one can prove Go\"dels theorem as a semi-trivial
> > consequence of the halting problem. so perhaps we should throw in
> > set theory and arithmetic as well.
>
>Could I see this proof.
>
> > Did I miss any of the classic ones?
>
>Classic representations of computation? Not as far as I can recall.
>
>--
>
>that is all
Edwin Young
> 2. Living things do not have on/off switches or reset buttons.
> They cannot be paused (e.g, by powerfail or checkpoint
> mechanisms) and then restarted.
> Computer simluations can. How, then, are computer simulations
> alive?
I'm not necessarily disagreeing with you, but isn't it possible to freeze
simple lifeforms and thaw them out later? This isn't quite the same as
pausing a computer (there's usually some damage), but it's a little
similar.
--
Edwin Young
Jonathan Stone
|> It's sort of an apples and oranges comparison. If Church's thesis is
|> correct (and the nifty thing is that if it is you can't prove it) there
|> ain't no such animal. Saying that we can find a way to compute
|> functions which are not recursively enumerable is like saying that we
|> can find a way to travel faster than light. Yeah, maybe, but that's not
|> the way to bet.
Thanks, Richard. That says, very clearly, pretty much what I was
thinking (and failed miserably to express).
P != (?) NP might be an even more apt comparison.
Jonathan Stone
Yup. Good point. its similar. But is it really the same?
Someone already cited a 60s paper about dumping larvae into liquid
helium. I forget what the otehr extreme was-- 3 kelvin to around 370?
And farmers routinely use artificial insemination: sperm packets
stored and shipped in liquid nitrogen.
But the immediate difference I see is this: Matt has agreed that
computer simulations are not "alive" when the computer is halted, or
paused, or what-have-you. (there are a few more possibilities,
but I'm saving them for later.)
And I'm assuming, tacitly, that biological organisms don't go from
being "alive" to being "not alive" (dead? suspended animation?), and
back to being "alive". Or that they can do this at any time in their
life. And you say this applies to "simple" life forms. But for
computer simulations, there's no real bound to the complexity that can
be checkpionted and later restored. (or copied Or reverted to. Both of
which are significant differences).
And simulations can do this with, literally, zero loss of
information-- which translates loosely, thermodynamically speaking,
to zero gain in entropy. to put it another way, ``suspended animation''
of simulations can last, in principle, forever. (Yes, sure, the
original hardware might wear out, but the nice thing about
Turing-completenes is that you can just emulate the old hardware
indefinitely.)
Think about the complexity issue and one perennial topic here:
panspermy. It's very hard to dow ith multicelled organisms. But with
software, it just means adding enough Reed-Solomon coding bits to
ensure that you can recover the data at the end of the flight.
Then there's exact duplication --- not merely a clone or identical
twin -- but another copy of the same individual with identical
environmental characteristics, memory, what-have-you. Or you could go
back to an earlier backed-up state, and restoring it.
Didn't C.J. Cherryh write a story where the principal (alien)
characters was an `ecology' of several different generations of
backups (software simluations) of the stored personalities of the
original alien crew? (yes, i used the word ecology, but in quotes. I
agree that its a convenient metaphor; but thats all it is).
Jonathan Stone
|> Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
|> > >> This may explain some of the differences here. To a computer
|> > >> scientist or mathematician, a 'computational element' is a device
|> > >> of sophisticated complexity to be able to model a Turing machine,
|> > >> and therefore able to "compute."
|> >
|> > > Nonsense.
|> >
|> >Well that was a useful remark. Would you care to do more than dismiss
|> >the entire mathematical basis of computation with a single word?
|>
|> The nonsense lies in saying that to a computer scientist or a
|> mathematician a computational element *is* a device et cetera. "A
|> computational element" isn't a standard term in formal treatments. I
|> don't doubt that someone uses the way you did - people are always
|> inventing neologisms. None-the-less. You follow this with an
|> egregious non sequitor, to wit your interpretation of "nonsense" as
|> "dismiss the entire mathematical basis of computation".
I think Marty's term of the use `computational element''
is well-established in the parallel programming community.
Try a PODC proceedings. Same goes for `node' and `network'
or the PRAM model. It seemed meaning seemed perfectly clear to me.
In CS, a "computational element" isnt a gate; we use "logic element"
or "gate" or "combinatorial logic" (as opposed to clocked logic,
like a latch) instead.
Given that, Marty's comment is not a non-sequitur at all. He defined
his terms and appealed to Turing-completeness; you said "nonsense".
Now, whether non-computational mathematicians use the term that way,
i have no clue. The process algebraists I hang out with seem
to follow it just fine.
Perhaps the difference in terminology is because CS goes beyond
formal treatments, into engineering and experimental "science"?
Anyway, I still cant see what's nonsense about Marty's statement, at
least if we leave mathematicians out of it. Is it purely terminological
differences? If not, could you try and say it again, please?
Jonathan Stone
[snip]
|> > You can have amazing machines based upon the crap buses, small
|> > cache and low clock rates of today - in theory.
|>
|> No, actually, you can't. If you don't fix the communication
|> bottlenecks, the CPUS sit there and wait a lot of the time. When you
|> do fix the hardware bottlenecks you end up with software communication
|> bottlenecks. When you finally balance the system, you discover that
|> most of your problems don't have the inherent parallism to take
|> advantage of it.
I hope this doesn tcome across the wrong way but: how long did it ake
you to figure that out? It's a really, really good summation.
The only thing it leaves out is this: many problems dont even have
enough parallellism to keep the multiple units in a *single* modern
CPU usefully busy.
And with the compute/communicate tradeoffs, thats just going
to get worse and worse.
Jonathan Stone
|> >>>>> Jonathan Stone pounded silicon into:
|> A couple of hours with my first computer architecture text. I can't
|> remember the name any more, but it was a classic. (Newell?) It
|> actually described the ways in which this becomes true.
Bell and Newell? (second edition was Bell, Siewioriek, and
Newell). McGraw-Hill. Forget the date. I cited it here to
someone a couple of weeks back.
|> A lot of the stuff we knew in the '70s got lost in the '80s because
|> the RiSC camp thought they could do no wrong and also thought they
|> didn't need any of the 'old' knowledge because they were 'reinventing'
|> computing.
Well, similarities between the CDC and RISC was widely noticed.
Just what, exactly, do you think got truly lost?
|> BTW, the vast majority of the parallel programming community *still*
|> doens't understand the distinction between 'cooperative' and
|> 'competitive multitasking, a couple of terms I tried to get into the
|> literature about 15 years ago.
There's metrics of how much communcation is needed. But the theorists
worked so long with PRAM models where communication is free. Say what
you like about RISC, you have to give Dave Patterson some credit for
standing up and saying that the PRAM model had to go.
|> > The only thing it leaves out is this: many problems dont even have
|> > enough parallellism to keep the multiple units in a *single*
|> > modern CPU usefully busy.
|>
|> Yup. wait until people see the delivered performance of Merced.
Yes, precious, gollum gollum. the trad rags already have people
waiting for mckinley.
Enough. we should take this to alt.folklore.computers or something.
Stew Dean
Stone) wrote:
>In article <36b4ff41...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
>|> In talk.origins I read this message from
>|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>|> Anyway,
>|> while there is some of content here, the bulk of this post is simply
>|> personal comments and I see no reason to try to pick out the little
>|> bit of content and respond. If anyone else wishes to argue these
>|> points, please repeat them.
>
>Nonsense. the absic content is that you dont know what you're talking
>about. Your comments make are a laughing-stock.
>
Insult.
>|>Jonathan wishes to repost with just the
>|> content, I will respond.
>
>Matt, you just dont get it. My basic point is that you are profoundly
>ignorant here: you simply dont know what you're talking about.
>
Insult
>My second point is that you are so ignorant you dont even appreciate
>just how foolish your comments are.
>
Insult
>My third point is that i'm fed up with *you* asking *me* to correct
>your mistakes. You make claims. The onus is on you to make them
>coherent and non-self-contradictory. But you don't.
Personal attack.
Content = 0
I think that proves Matt's point. Your lack of control does not help
any points you do have (none in this case). As with any replies saying
simply 'you're wrong' or using personal attacks it only helps confirm
the point made by the person being attacked or told 'they're wrong
because I said so' .
I say this only in the hope we can keep these debates open and
interesting.
Cheers
Stewart Dean - ste...@webslave.dircon.co.uk
alife guide - http://www.webslave.dircon.co.uk/alife
Stew Dean
Stone) wrote:
>[posted, d-and hopefully, emailed]
>The only real difference is time. If you want to decide that life
>depends on the rate at which creatures "live", then (again) you are
>profoundly ignorant, since even on earth, metabolic rates vary with
>temperature. And besides, your argument was based on _complexity_,
>not rates of state change associated with that complexity.
>
Complexity is based on rates of state change! This is what mostly
defines a complex system as opposed to one where rate is too high
(chaotic) or to low (periodic and static). The rate of change is not
about time taken but steps of the process in relation to each other,
usually the bigger a system the longer this the rate of change can be.
>
>You've been a thorough ignoramus in this whole arugment, Matt.
You're walking on very thin ice Jonathan. Got a feeling who ever
taught you is a decade or so out of date.
Jonathan Stone
|> On 31 Jan 1999 19:16:28 -0500, jona...@Cup.DSG.Stanford.EDU (Jonathan
|> Stone) wrote:
|>
|> >[posted, d-and hopefully, emailed]
|>
|> >The only real difference is time. If you want to decide that life
|> >depends on the rate at which creatures "live", then (again) you are
|> >profoundly ignorant, since even on earth, metabolic rates vary with
|> >temperature. And besides, your argument was based on _complexity_,
|> >not rates of state change associated with that complexity.
|> >
|> Complexity is based on rates of state change!
No, it isn't. End of story.
Jonathan Stone
|> > What constitutes a node and what kind of systems are you talking
|> > about?
|>
|> I'm talking about a massively parallel processor. That is the kind of
|> computer that people think the brain is a good model for the
|> construction of. IBM and Intel both have designed and built machines
|> of the size I'm describing.
[SP2 and paragon? or the nwere IBM box?]
Or Maspar, CM-5, or the Cray T3D/T3E. Or the old IBM SP2.
Richard, he means a CPU and some amount of local memory and some fast,
lowlatency communcation to two or more CPUs. Where low-latency means
only two orders of magnitude slower to access a remote node's memory
versus a local one. ``NUMA'' is the current buzzword; also `massively
parallel'.
|> > What are you counting as a logic element?
|>
|> anything capable of modeling a boolean function, plus any memory
|> cells. I assume that's what you meant by 'logic element', as it is
|> what a EE would mean.
Yup.
|> > I will cheerfully concede that if you network enough machines
|> > together you can accumulate enough components to match the neuron
|> > count of a single human brain. It remains, though, that a single
|> > machine has orders of magnitude fewer components than the brain.
|> > (The difficulties of comparing components are also cheerfully
|> > conceded.)
|>
|> You confuse desktop computers with 'machines'. The Intel Paragon and
|> the IBM GF-10 are considered by computer architects to be single
|> machines, and, when properly configured, easily exceed 100M 'logic
|> elements'.
Yup. Marty's nomenclature is, again, standard. And the communication
network linking these machines is not the same kind of beast as a
commodity LAN, either.
Jonathan Stone
|> On 31 Jan 1999 20:54:44 -0500, jona...@Cup.DSG.Stanford.EDU (Jonathan
|> Stone) wrote:
|> Content = 0
|>
|> I think that proves Matt's point. Your lack of control does not help
|> any points you do have (none in this case). As with any replies saying
|> simply 'you're wrong' or using personal attacks it only helps confirm
|> the point made by the person being attacked or told 'they're wrong
|> because I said so' .
|>
|> I say this only in the hope we can keep these debates open and
|> interesting.
Um, no, I dont agree. the problem is that Matt really *doesnt* know
what he's talking about.
Stew Dean
Stone) wrote:
>In article <36b7ffd7...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
>|> In talk.origins I read this message from
>|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>|> Jonathan, I am glad you have these anonymous authorities on your side.
>
>Matt, you are a hypocrite. You have appealed to the authority of
>un-named personal acquaintances yourself. But If you want the names
>of my friends, or their relevaent qualifications, you'll get them.
>Provided my friends agree. (at least one of them woudlnt want to give
>their details to someone who says the stupid things you've been saying.)
I think they should take your place here. I'd be interested in their
views. You appear to be more interested in chest beating.
>Not that that makes what you've been saying any less stupid.
>Your statements stand on their own grounds.
So disagree without the use of appeals to authority, unsupported
'you're wrongs' and insults of 'you know nothing'.
Go to the library - get a book on debating or reasoning or philosophy
and hopefully you'll be able to make your points clearer.
PZ Myers
[snip]
>>The estimate that human brains contain three orders of magnitude more
>>of these 'logic circuit' equivalents is off by quiet a bit, then. The
>>largest extent computer systems have > 1000 nodes where each node has
>>> 100,000,000 'logic elements' which would put it on a par with a
>>human brain.
>
>I'm not sure what you are counting here. What constitutes a node and
>what kind of systems are you talking about? What are you counting as a
>logic element? I will cheerfully concede that if you network enough
>machines together you can accumulate enough components to match the
>neuron count of a single human brain. It remains, though, that a single
>machine has orders of magnitude fewer components than the brain. (The
>difficulties of comparing components are also cheerfully conceded.)
It's also very difficult to count the number of elements in a human brain.
I've seen estimates as high as 10**14 for the number of neurons -- 10**12
is a more typical number. Each neuron receives, on average, roughly 10**3
synapses and makes that same number. If a 'connection' is equivalent to
a logic element, that puts us at about 10**15 already.
However, even if the computer scientists are getting up into the ballpark
in terms of number, they must be putting the pieces together all wrong.
I don't see any computers that exhibit the degree of flexibility and
complexity of output demonstrated by even a fly brain, with 10**5 neurons.
Does anyone believe any computer could simulate the complexity involved
in sorting out multiple sensory modalities to navigate and operate a
complex flying machine to a piece of rotting fruit, where it will land,
eat, groom, seek out members of the opposite sex of the same species,
go through elaborate mating rituals, mate, and reproduce? I don't think so.
--
PZ Myers
Jonathan Stone
|> In article <36b682cc...@news.dircon.co.uk>, stew...@webslave.dircon.co.uk (Stew Dean) writes:
|> |> On 31 Jan 1999 19:16:28 -0500, jona...@Cup.DSG.Stanford.EDU (Jonathan
|> |> Stone) wrote:
|> |>
|> |> >[posted, d-and hopefully, emailed]
|> |>
|> |> >The only real difference is time. If you want to decide that life
|> |> >depends on the rate at which creatures "live", then (again) you are
|> |> >profoundly ignorant, since even on earth, metabolic rates vary with
|> |> >temperature. And besides, your argument was based on _complexity_,
|> |> >not rates of state change associated with that complexity.
|> |> >
|> |> Complexity is based on rates of state change!
|>
|> No, it isn't. End of story.
Or, if you want a more substantive answer:
As long as we're talking about *computation*, then complexity *cannot*
have anything to do with ``rates of change'', since formal models of
computation do not includes rates of change. and by the Church-Turing
thesis, all models of computation are the same.
As computations go, pencil-and-paper computation is, by definition,
*the same* computation as on the fastest computer you can buy.
Since in both cases the algorithms are *the same*, their complexity
is, immediately, the same. Thats almost *the* basic fundamental.
You want references?
Complexity could be defined according to Komolgorov (or Komologorov/Chaitin)
complexity. Thats sort-of analogoius to Shannon/weaver information theory.
Or it could be big-Oh time/space bounds. But neither of those says
anything at all about the *rate* of computation.
I have no idea what you're talking about, but I dont see how it has
any bearing at all to the complexity of any computation.
|> This is what mostly
|> defines a complex system as opposed to one where rate is too high
|>(chaotic) or to low (periodic and static). The rate of change is not
|> about time taken but steps of the process in relation to each other,
|> usually the bigger a system the longer this the rate of change can be.
Um, just precisely *how* does mean a manual, pencil-and-paper
execution of an algorithm, has a different complexity than a
high-speed digital computer evaluation of that *same* algorithm?
Thats what I asked matt. He said one was alive, and the other wasn't.
Yet even by your metric, from what I can see (looking at what you say
about `steps of processes in relation to each other', the
pencil-and-paper and the digital computer do indeed have the same
complexity.
| >>You've been a thorough ignoramus in this whole arugment, Matt.
>You're walking on very thin ice Jonathan. Got a feeling who ever
>taught you is a decade or so out of date.
You're walking on very thin ice yourself, Stewart.
Jonathan Stone
|> In article <36b55319...@199.0.65.59>, c...@tiac.net (Richard Harter) wrote:
|>
|> [snip]
[snip]
|> However, even if the computer scientists are getting up into the ballpark
|> in terms of number, they must be putting the pieces together all wrong.
No, Paul, they're putting the pieces together very well indeed, given
that the costs of connections versus `compute' (neuron-level in VLSI
differs by orders of magnitude from the same traedoff with (biological)
neurons.
You want a 2-D mesh in silicon? fine. 3-d? Come back when we can
build photonic computing elements.
|> Does anyone believe any computer could simulate the complexity involved
|> in sorting out multiple sensory modalities to navigate and operate a
|> complex flying machine to a piece of rotting fruit, where it will land,
|> eat, groom, seek out members of the opposite sex of the same species,
|> go through elaborate mating rituals, mate, and reproduce? I don't think so.
Why not? I dont see any reason in principle why not. I think the
fundamental problem is that we dont really know how *flies* do it.
If we knew how flies did it, if we had a mathematical model,
then we could make computers could do it, too.
Or are you betting against the Church-Turing thesis, too?
If so, I'd really really like to hear why.
Jonathan Stone
|> On 31 Jan 1999 21:01:32 -0500, jona...@Cup.DSG.Stanford.EDU (Jonathan
|> Stone) wrote:
|>
|> >In article <36b7ffd7...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
|> >|> In talk.origins I read this message from
|> >|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|>
|> >|> Jonathan, I am glad you have these anonymous authorities on your side.
|> >
|> >Matt, you are a hypocrite. You have appealed to the authority of
|> >un-named personal acquaintances yourself. But If you want the names
|> >of my friends, or their relevaent qualifications, you'll get them.
|> >Provided my friends agree. (at least one of them woudlnt want to give
|> >their details to someone who says the stupid things you've been saying.)
|>
|> I think they should take your place here. I'd be interested in their
|> views. You appear to be more interested in chest beating.
|>
|> >Not that that makes what you've been saying any less stupid.
|> >Your statements stand on their own grounds.
|>
|> So disagree without the use of appeals to authority, unsupported
|> 'you're wrongs' and insults of 'you know nothing'.
I never said Matt knows nothing. He knows more about the nuts and
bolts of biology than I do. And if you would care to look at
DejaNews, I think you will find note that Matt started the appeals to
un-named, anonymous authority, when he claimed that, in mathematics, a
false statement is really true, if you take it as an axiom.
Matt has made false assertions about what popularizations say, in ways
that sugsest he doesnt understand them. He's been offered citations
to standard textbooks in the field (Enderton's text on formal logic
and the foundations of mathematics), and he pooh-poohed them.
|> Go to the library - get a book on debating or reasoning or philosophy
|> and hopefully you'll be able to make your points clearer.
How much clearer can I get than posting the same questions over and
over and over again? All that happens is that Matt decline to answer
them, or insults me, or tries and get me to answer them for him.
Matt says, in effect, he's right unless I show he's wrong.
Even when I show hes' wrong, he doesnt accept it.
And you think I should take lessons in reasoning?
Jonathan Stone
|>
|> >>>>> Tim Tyler pounded silicon into:
|>
|> [snip]
|>
|> > See, for one example, the *advocate* of machine intelligence Hans
|> > Moravec's recent 'Robot' book. By his metrics, desktop computers
|> > are currently somewhere around the cockroach level in terms of
|> > processing power.
|>
|> > http://www.frc.ri.cmu.edu/~hpm/book98/ has the graphs, IIRC.
|>
|> > Human brains massively out-perform modern computers in a wide
|> > range of real-world tasks. Their superior, massively parallel
|> > connectionist architecture probably helps them to do this. Why do
|> > you think the neural net people spend so much time trying to copy
|> > nature's work?
|>
|> I think they are doing it because they don't have the necessary
|> grounding in the philosophy of mathematics or the history of
|> mathematics to realize that digital computers are limited mathematical
|> tools.
Perhaps the same applies to advocates of artificial life?
Richard Harter
>But the immediate difference I see is this: Matt has agreed that
>computer simulations are not "alive" when the computer is halted, or
>paused, or what-have-you. (there are a few more possibilities,
>but I'm saving them for later.)
The general problem with the argument you guys are having is in the
unclearness of the term "alive". Your argument really amounts to
pointing out that simulations are not alive in the way that "life as we
know it" is. I think (and I'm not sure) that Matt is taking a view that
the essential features of life are captured in (appropriate) computer
simulations and possibly other things. (IIRC he has advance the view
that IBM - the corporation - is alive.) There is nothing wrong with
this per se - it is simply extending the notion of what it means to be
"alive" - but the burden is on Matt to clarify what "alive" means in his
sense of the word.
Richard Harter
>In article <wksocq8...@usenet.nospam.fogey.com>, Marty Fouts <usene...@usenet.nospam.fogey.com> writes:
>
>
>|> > What constitutes a node and what kind of systems are you talking
>|> > about?
>|>
>|> I'm talking about a massively parallel processor. That is the kind of
>|> computer that people think the brain is a good model for the
>|> construction of. IBM and Intel both have designed and built machines
>|> of the size I'm describing.
>[SP2 and paragon? or the nwere IBM box?]
>
>Or Maspar, CM-5, or the Cray T3D/T3E. Or the old IBM SP2.
>
>Richard, he means a CPU and some amount of local memory and some fast,
>lowlatency communcation to two or more CPUs. Where low-latency means
>only two orders of magnitude slower to access a remote node's memory
>versus a local one. ``NUMA'' is the current buzzword; also `massively
>parallel'.
>|> > What are you counting as a logic element?
>|>
>|> anything capable of modeling a boolean function, plus any memory
>|> cells. I assume that's what you meant by 'logic element', as it is
>|> what a EE would mean.
>
>Yup.
That's what I assumed but I was just checking.
>|> > I will cheerfully concede that if you network enough machines
>|> > together you can accumulate enough components to match the neuron
>|> > count of a single human brain. It remains, though, that a single
>|> > machine has orders of magnitude fewer components than the brain.
>|> > (The difficulties of comparing components are also cheerfully
>|> > conceded.)
>|>
>|> You confuse desktop computers with 'machines'. The Intel Paragon and
>|> the IBM GF-10 are considered by computer architects to be single
>|> machines, and, when properly configured, easily exceed 100M 'logic
>|> elements'.
>Yup. Marty's nomenclature is, again, standard. And the communication
>network linking these machines is not the same kind of beast as a
>commodity LAN, either.
Let me see if I have this right. There are networks consisting of ~1000
nodes where each node is an Intel Paragon or equivalent having ~100M
logic elements? The machine size doesn't surprise me but the notion of
1000 such machines networked together does. Who on Earth uses such a
network? Are we talking about an integrated system here?
Jonathan Stone
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
|>
|>
|> >But the immediate difference I see is this: Matt has agreed that
|> >computer simulations are not "alive" when the computer is halted, or
|> >paused, or what-have-you. (there are a few more possibilities,
|> >but I'm saving them for later.)
|>
|> The general problem with the argument you guys are having is in the
|> unclearness of the term "alive".
Yup. thats part of it.
|> Your argument really amounts to
|> pointing out that simulations are not alive in the way that "life as we
|> know it" is.
Yup. Thats part of it.
|> I think (and I'm not sure) that Matt is taking a view that
|> the essential features of life are captured in (appropriate) computer
|> simulations and possibly other things. (IIRC he has advance the view
|> that IBM - the corporation - is alive.)
Yes. This is the one that caused my biolgist friends to burst out
laughing. When I showed them Matt's message and saw that he meant it
*seriously*, they stared gape-jawed and told me to stop wasting time
arguing with such idiots.
But who knows, perhaps my friends and i just share the same biases.
|> There is nothing wrong with
|> this per se - it is simply extending the notion of what it means to be
|> "alive" - but the burden is on Matt to clarify what "alive" means in his
|> sense of the word.
Yes. Precisely. That is the second-biggest part of what the
disagreement is about. I keep asking difficult questions, in the hope
of getting Matt to either clarify what he means, such that I or some
other skeptic could apply his sense; or retract it.
So far, neither has happened.
i'd add one more thing:
I think Matt also needs to formulate his concept of "life" in ways
that dont contradict other branches of sciene, or mathematics. So far,
Matt's sense of how simulatinos are "alive" violates the Church-Turing
thesis: fast digital-copmuter simulations are "alive", but the *exact*
*same* computation, done with pencil and paper, isnt "alive". (i'd
guess that violates the K/C sense of "information", too, but i havent
asked hard questions about that.)
In my book, thats another sequence-heme level blooper.
Richard Harter
>In article <7940v1$ov4$9...@nntp.Stanford.EDU>, jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) writes:
>|> In article <36b682cc...@news.dircon.co.uk>, stew...@webslave.dircon.co.uk (Stew Dean) writes:
>|> |> Complexity is based on rates of state change!
>|>
>|> No, it isn't. End of story.
>
>Or, if you want a more substantive answer:
>
>As long as we're talking about *computation*, then complexity *cannot*
>have anything to do with ``rates of change'', since formal models of
>computation do not includes rates of change. and by the Church-Turing
>thesis, all models of computation are the same.
>
>As computations go, pencil-and-paper computation is, by definition,
>*the same* computation as on the fastest computer you can buy.
>
>Since in both cases the algorithms are *the same*, their complexity
>is, immediately, the same. Thats almost *the* basic fundamental.
I suspect that he is talking about finite-state cell networks as in life
(Conway's game of life) - rate of change is the percentage of cells that
change state in each step. Stasis/complexity/chaos is the tip off.
There are guys at Santa Fe who have developed a complexity measure for
this sort of thing. The basic idea matches our intuitive sense of what
is meant by complexity better than Kolmorgorov complexity does.
Jonathan Stone
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
|> Let me see if I have this right. There are networks consisting of ~1000
|> nodes where each node is an Intel Paragon or equivalent having ~100M
|> logic elements?
Nope, the "Intel Paragon", or the IBM system Marty mentioned, or a
large Cray T3E, is the whole shebang. Each `Node' is a CPU plus local
memory, or a small-scale multi-CPU module. There are ~1000 such
nodes, or CPUs, in the system.
|. The machine size doesn't surprise me but the notion of
|> 1000 such machines networked together does.
The usual suspects: traditional supercomputer customers.
|> Who on Earth uses such a
|> network? Are we talking about an integrated system here?
to oversimplify drastially:
Yes. Not a LAN, nothing like it. its a special-purpose network that
passes around messages. Think of a commodity intel ddesktop
motherboard with slots for 4 or 8 CPUs. That motherboard includes a
network that ties the separate CPUs together.
We're talking about machines with a mesh of rougly comparable
interconnects, set up in a switching frabric between each of the ~1000
CPU boards. Parallel programs communicate by sending `messages'
across that interconnect.
Um. If memory serves, the original 'Paragon' was a backplane node that
supported, um, a total of 32 CPU or memory boards. Mayeb that's what
Marty means by a "node". Its been a long time, though.
Richard Harter
In principle, yes. In practice, maybe, maybe not. Analog, remember.
Neurons are cells with all of the chemical complexity of cells.
Moreover the number of neurons is far from the whole story. Brains,
even fly brains, require bodies.
Jonathan Stone
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
|>
|> >|> In article <36b682cc...@news.dircon.co.uk>, stew...@webslave.dircon.co.uk (Stew Dean) writes:
[snip]
|> >As long as we're talking about *computation*, then complexity *cannot*
|> >have anything to do with ``rates of change'', since formal models of
|> >computation do not includes rates of change. and by the Church-Turing
|> >thesis, all models of computation are the same.
|> >
|> >As computations go, pencil-and-paper computation is, by definition,
|> >*the same* computation as on the fastest computer you can buy.
|> >
|> >Since in both cases the algorithms are *the same*, their complexity
|> >is, immediately, the same. Thats almost *the* basic fundamental.
|>
|> I suspect that he is talking about finite-state cell networks as in life
|> (Conway's game of life) - rate of change is the percentage of cells that
|> change state in each step.
But I'm *not* talking about cellular automata, and neither was matt,
that i can tell.
|> Stasis/complexity/chaos is the tip off.
Sorry, to me that's just a tip-off for fuzzy new-age thinking, and I
answered it accordingly. Celluare automata are just one more
turing-equivalent formalism, and not a very interesting one at that.
|> There are guys at Santa Fe who have developed a complexity measure
|> for this sort of thing. The basic idea matches our intuitive sense of
|> what is meant by complexity better than Kolmorgorov complexity does.
Kolmorogov or big-Oh does just fine for me, thanks.
But could you explain just how it has any bearing at *all* on why a
fast digital-computer simulation is alive, while the same
pencil-and-paper computation isn't? Take the Santa-Fe metric of
complexity, and run Matt's simulation on a UTM coded as a cellular
automaton on (a) a fast digital computer, and (b) pencil and paper.
My reductio-ad-adsurbum argument is identical in both cases.
So either way, Matt's "life" violates Turning equivalence.
So, I think Stewart Dean (a) didn't make it at all clear what he was
talking about (b) didnt read what i wrote carefully enough, (c) has
only strengthened that particualr argument against "life" sensu Matt.
But if you see it differently, I'd be interested to hear it.
Richard Harter
>In article <36b59bc4...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
>|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
>
>|> Let me see if I have this right. There are networks consisting of ~1000
>|> nodes where each node is an Intel Paragon or equivalent having ~100M
>|> logic elements?
>
>Nope, the "Intel Paragon", or the IBM system Marty mentioned, or a
>large Cray T3E, is the whole shebang. Each `Node' is a CPU plus local
>memory, or a small-scale multi-CPU module. There are ~1000 such
>nodes, or CPUs, in the system.
Okay. Paragon is the name of the whole machine. Then the statement is
that each CPU has ~100M elements. Am I correct in assuming that the
nodes consist mostly of parallel processing arrays?
>|. The machine size doesn't surprise me but the notion of
>|> 1000 such machines networked together does.
>
>The usual suspects: traditional supercomputer customers.
Oh yes, if it's a single machine, I know who they are. I'm acquainted
with that class of architecture but that was quite some time ago;
they've gotten bigger than I'd realized.
>|> Who on Earth uses such a
>|> network? Are we talking about an integrated system here?
>
>to oversimplify drastially:
>
>Yes. Not a LAN, nothing like it. its a special-purpose network that
>passes around messages. Think of a commodity intel ddesktop
>motherboard with slots for 4 or 8 CPUs. That motherboard includes a
>network that ties the separate CPUs together.
>
>We're talking about machines with a mesh of rougly comparable
>interconnects, set up in a switching frabric between each of the ~1000
>CPU boards. Parallel programs communicate by sending `messages'
>across that interconnect.
>
>Um. If memory serves, the original 'Paragon' was a backplane node that
>supported, um, a total of 32 CPU or memory boards. Mayeb that's what
>Marty means by a "node". Its been a long time, though.
>
Richard Harter, c...@tiac.net, The Concord Research Institute
Richard Harter
>In article <36b5a22a...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
>|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
>|>
>|> >In article <7940v1$ov4$9...@nntp.Stanford.EDU>, jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) writes:
>|> >|> In article <36b682cc...@news.dircon.co.uk>, stew...@webslave.dircon.co.uk (Stew Dean) writes:
>
>[snip]
>
>|> >As long as we're talking about *computation*, then complexity *cannot*
>|> >have anything to do with ``rates of change'', since formal models of
>|> >computation do not includes rates of change. and by the Church-Turing
>|> >thesis, all models of computation are the same.
>|> >
>|> >As computations go, pencil-and-paper computation is, by definition,
>|> >*the same* computation as on the fastest computer you can buy.
>|> >
>|> >Since in both cases the algorithms are *the same*, their complexity
>|> >is, immediately, the same. Thats almost *the* basic fundamental.
>|>
>|> I suspect that he is talking about finite-state cell networks as in life
>|> (Conway's game of life) - rate of change is the percentage of cells that
>|> change state in each step.
>
>But I'm *not* talking about cellular automata, and neither was matt,
>that i can tell.
>|> Stasis/complexity/chaos is the tip off.
>
>Sorry, to me that's just a tip-off for fuzzy new-age thinking, and I
>answered it accordingly. Celluare automata are just one more
>turing-equivalent formalism, and not a very interesting one at that.
Not very interesting to you, perhaps.
>|> There are guys at Santa Fe who have developed a complexity measure
>|> for this sort of thing. The basic idea matches our intuitive sense of
>|> what is meant by complexity better than Kolmorgorov complexity does.
>
>Kolmorogov or big-Oh does just fine for me, thanks.
They're fine for computational purposes but not for characterizing
structure.
>But could you explain just how it has any bearing at *all* on why a
>fast digital-computer simulation is alive, while the same
>pencil-and-paper computation isn't?
No bearing at all, as far as I can tell.
>Take the Santa-Fe metric of
>complexity, and run Matt's simulation on a UTM coded as a cellular
>automaton on (a) a fast digital computer, and (b) pencil and paper.
>My reductio-ad-adsurbum argument is identical in both cases.
I've lost track of what reductio you're talking about.
>So either way, Matt's "life" violates Turning equivalence.
>
>
>So, I think Stewart Dean (a) didn't make it at all clear what he was
>talking about (b) didnt read what i wrote carefully enough, (c) has
>only strengthened that particualr argument against "life" sensu Matt.
(a) and (b) I agree with, (c) I don't know about, having lost track.
>But if you see it differently, I'd be interested to hear it.
If we're talking about paper and pencil vs machine the only real
difference is that, as a practical matter, we can't do p&p on the
requisite scale. Otherwise the formal structure is the same.
Richard Harter
>>>>>> Richard Harter pounded silicon into:
>
> > jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
>
> >> In article <36b59bc4...@199.0.65.59>, c...@tiac.net (Richard
> >> Harter) writes: |> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone)
> >> wrote:
> >>
> >> |> Let me see if I have this right. There are networks
> >> consisting of ~1000 |> nodes where each node is an Intel Paragon
> >> or equivalent having ~100M |> logic elements?
> >>
> >> Nope, the "Intel Paragon", or the IBM system Marty mentioned, or
> >> a large Cray T3E, is the whole shebang. Each `Node' is a CPU plus
> >> local memory, or a small-scale multi-CPU module. There are ~1000
> >> such nodes, or CPUs, in the system.
>
> > Okay. Paragon is the name of the whole machine. Then the
> > statement is that each CPU has ~100M elements. Am I correct in
> > assuming that the nodes consist mostly of parallel processing
> > arrays?
>
>Not the CPU, but the CPU + its local memory. The nodes mostly consist
>of memory parts, the CPUs are standard off-the-shelf intel processors.
That's a bit different. I thought you were talking about processors
with 100M elements. Color me much less impressed. Surely we're not
talking about standard type RAM here, or are we?
This doesn't put it in quite the same ballpark as the brain; the machine
has as many logic elements as a brain has neurons (actually 100G is a
lower estimate) but a neuron has considerably more functionality than a
simple logic element and the neuron does have that big fan-out.
>The GF11 was an earlier IBM design that had more discrete components.
>It was designed specifically to solve a particular problem in QCD.
>The SP-3 is IBM's equivalent to the paragon and Cray T3 that Jonathan
>is mentioning.
IIRC the early Cray's had extensive parallel processing arrays. Is that
no longer the case?
Matt Silberstein
jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
>In article <36b621aa...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
>
>This is getting ridicuolus. Here is the list of questions again,
>for anyone who hasn't noticed that Matt is consistently avoiding
>answering them:
>
Even though I will, in general avoid all of Jonathan's posts in
future, this has enough content to ad hominem ration to be worth
dealing with.
>
> 1. Matt's definitions of `life' are metaphorical.
> THe interpretation of `metabolize' and `reproduce' is so loose
> that, as i see it, it includes flames and crystals.
>
> Are flames and crystals alive?
I have answered this. Flames and crystals do have some of the
properties of life, but not much. On a scales they are quite close to
a non-crystal rock. I don't see life as a binary condition so I can't
answer yes/no.
Now maybe
>
> 2. Living things do not have on/off switches or reset buttons.
> They cannot be paused (e.g, by powerfail or checkpoint
> mechanisms) and then restarted.
> Computer simluations can. How, then, are computer simulations
> alive?
>
I have answered this several times as well. I consider the current
ability to turn off and on ALife a significant drawback. It shows that
the current candidates are not very alive. But I point out that
viruses can be turned off and on as well. I will repeat. Life seems
like a continuum. We have rocks on one end, cats on the other, and
viruses in the middle. Crystals and flames are close to rocks. Current
ALife is somewhere between a rock/crystal/flame and a virus.
> 3. Computer simulations do not acutally metabolize.
> The level of abstraction wehre we find similarties between
> a living thing and a computer simluation of a living
> thing does not actually exchange energy: instead, it
> exchanges information. (Any energy exchagne takes place
> purely within the computational devices themselves).
>
> Precisely how does exchange of information constitue
> *real* `metabolism' as opposed to a *smulation* of metabolism?
>
I am not sure why metabolism must necessarily mean energy (in a
physics sense) as opposed to meaning some sort of subsistence. You
want to dismiss this as "metaphorical", go ahead. I see it as
exploring the ideas.
> 4. Computer simluations are merely comptuational models.
> They can be executed on a huge range of computational devices,
> all the way from supercomputers to pencil and paper.
>
> Mathematically, these computations are all identical:
> that is the Church-Turing thesis.
>
> Is a computation done on pencil and paper alive?
>
> If not, then Matt's orignal claim is refuted; there
> is something to life *APART* from the complexity and
> a homomorphism between biological organisms and simluations
> of biological organisms.
>
Rate of action *in the real world* is important. If there was such a
situation with a person doing the work on paper and pencil it might, I
suppose, be alive. But I doubt any of us could possible recognize it.
A paper and pencil implementation of a 3-D graphics program may be
formally equivalent to the program running on a computer, but I doubt
a person could recognize the resultant output as the same.
> 5. Which goedel numbers of computations are alive, and which
> are not, and why?
>
Why do I need to have some mechanism for determining this? So some may
and may not be. Maybe we will figure out some wonderful way to make
this determination, or maybe we won't. So what?
> 6. Computer simulations are, by definition, mathematical models
> of the things being simulated. The models abstract away
> from the real thing. They deal only with some mathematical
> aspects of the thing being modelled.
> They are not the thing itself.
>
> For example, a simulation of a nuclear power-plant
> may model the nuclear reactions and the energy transfer,
> but the simulated energy cannot power a real light-bulb.
> Nor can a simulation of an aircraft acutally tranport people
> or things. ``the map is not the territory''.
>
> Matt's basic thesis is that, in the case of life,
> the map *is* the territory. This seems to me, and to Marty,
> to be a fundamental semantic error.
>
You assert that they are simulations and you your label as an
argument. The question is what are the important qualities of life.
Reproduction is process, ISTM it does not matter if the system is
based on DNA or something else. And parts of ALife systems do, after a
fashion, reproduce. They don't simulate reproduction, they do it. Now
maybe these systems do not have enough of the qualities of life and
maybe they never will. But that is an issue to discuss and explore,
not to assert. Arguing that tradition says one thing so we can't say
another does not fly. Nor does arguing that we have categorized the
systems are simulations so they are *only* simulations.
> What is so special about life that, unliek all other
> simluations, the map really *is* the territory?
>
>
>
>I've asked Matt these questions repeatedly. He has not answered any
>except 5 (and I still havent seen that). His only response is to ask
>*me* to answer the questions. Even though I've repeatedly, that I
>find that exceedingly offensive (because, as you can guess from the
>questions, I think Matt's idea is bloody stupid, and I dont *want* to
>try and fix it up.)
>
>It's really up to Matt. He can either try and answer the questions
>substantively, or not.
>
>We're all adult enough to see who is being intellecutally dishonest
>here, and who isn't.
Agreed.
Matt Silberstein
-------------------------------------------------------
Perfection (in design) is achieved not when there is nothing
more to add, but rather when there is nothing more to take away.
Antoine de Saint-Exupéry (who was an aviator and aircraft designer
when he wasn't being the author of classic children's books)
PZ Myers
>In article <myers-01029...@chestnut1-4.slip.netaxs.com>,
my...@netaxs.com (PZ Myers) writes:
>|> In article <36b55319...@199.0.65.59>, c...@tiac.net (Richard
Harter) wrote:
>|>
>|> [snip]
>
>[snip]
>
>|> However, even if the computer scientists are getting up into the ballpark
>|> in terms of number, they must be putting the pieces together all wrong.
>
>No, Paul, they're putting the pieces together very well indeed, given
>that the costs of connections versus `compute' (neuron-level in VLSI
>differs by orders of magnitude from the same traedoff with (biological)
>neurons.
They're putting them together very well to accomplish very tightly defined
tasks with a great deal of efficiency. That I would agree with.
However, organic brains have a kind of amazingly flexible, robust versatility
that is totally lacking in our computers. I don't think that just adding
more and more elements is going to get you any closer to that kind of
behavior (and maybe you don't want to -- computers and flies have rather
different jobs to do).
>
>You want a 2-D mesh in silicon? fine. 3-d? Come back when we can
>build photonic computing elements.
>
>
>
>|> Does anyone believe any computer could simulate the complexity involved
>|> in sorting out multiple sensory modalities to navigate and operate a
>|> complex flying machine to a piece of rotting fruit, where it will land,
>|> eat, groom, seek out members of the opposite sex of the same species,
>|> go through elaborate mating rituals, mate, and reproduce? I don't think so.
>
>Why not? I dont see any reason in principle why not. I think the
>fundamental problem is that we dont really know how *flies* do it.
>If we knew how flies did it, if we had a mathematical model,
>then we could make computers could do it, too.
>
>
>Or are you betting against the Church-Turing thesis, too?
>If so, I'd really really like to hear why.
Oh, I think *eventually* we might be able to produce something with the
abilities of a fly or a mouse or maybe even a human. I'm not arguing at
all against the possibilities.
What I am critical of is attempts to relate the computation power of an
organic brain with a computer by counting bits and pieces. You know there
is a lot more to performance in a computer than the number of transistors
on a chip, and it is the same with brains. Flies have 10**5 neurons, which
doesn't sound like much if you compare it to the number of transistors on
even a 10-year-old CPU design. However, it is pretty obvious that flies
have much more sophisticated processors than that number implies. I think
that the number 10**12 is a similar understatement for the computing power
of a human brain, and no computer anywhere comes even close, yet.
--
PZ Myers
PZ Myers
<usene...@usenet.nospam.fogey.com> wrote:
>>>>>> PZ Myers pounded silicon into:
>
> > In article <36b55319...@199.0.65.59>, c...@tiac.net (Richard
> > Harter) wrote:
>
> > [snip]
>
> >>> The estimate that human brains contain three orders of magnitude
> >>> more of these 'logic circuit' equivalents is off by quiet a bit,
> >>> then. The largest extent computer systems have > 1000 nodes
> >>> where each node has
> >>>> 100,000,000 'logic elements' which would put it on a par with a
> >>> human brain.
> >> I'm not sure what you are counting here. What constitutes a
> >> node and what kind of systems are you talking about? What are
> >> you counting as a logic element? I will cheerfully concede that
> >> if you network enough machines together you can accumulate enough
> >> components to match the neuron count of a single human brain. It
> >> remains, though, that a single machine has orders of magnitude
> >> fewer components than the brain. (The difficulties of comparing
> >> components are also cheerfully conceded.)
>
> > It's also very difficult to count the number of elements in a
> > human brain. I've seen estimates as high as 10**14 for the number
> > of neurons -- 10**12 is a more typical number. Each neuron
> > receives, on average, roughly 10**3 synapses and makes that same
> > number. If a 'connection' is equivalent to a logic element, that
> > puts us at about 10**15 already.
>
> > However, even if the computer scientists are getting up into the
> > ballpark in terms of number, they must be putting the pieces
> > together all wrong.
>
>Um, before we go getting all judgemental, let us try to remember that
>digital computers serve a different purpose than human brains. What
>the designers of computer systems do isn't 'wrong', it just has to
>meet different requirements.
Yes, exactly. I wasn't trying to be judgemental.
>
> > I don't see any computers that exhibit the degree of flexibility
> > and complexity of output demonstrated by even a fly brain, with
> > 10**5 neurons.
>
>Agreed. There's somebody at MIT whose name I can never recall who took
>this observation to heart and has a whole different approach to robot
>building that can be viewed, in AI terms, as 'bottom up' rather than
>the usual 'bottom down' approach of the AI community.
Yes, I've seen those.
>
> > Does anyone believe any computer could simulate the complexity
> > involved in sorting out multiple sensory modalities to navigate
> > and operate a complex flying machine to a piece of rotting fruit,
> > where it will land, eat, groom, seek out members of the opposite
> > sex of the same species, go through elaborate mating rituals,
> > mate, and reproduce?
>
>marvin minsky does. so do most ai researchers, and, apparently now,
>the artificial life fadists.
>
> > I don't think so.
>
>me either.
Although I would qualify it a bit to say they can't do it now, or in the
immediately foreseeable future. But it ought to be possible. Of course, it
would mean abandoning many of the pragmatic goals of computer design to
make something that maybe can't add two numbers together very quickly,
but does something else without immediate applicability quite well.
--
PZ Myers
Richard Harter
>>>>>> Richard Harter pounded silicon into:
>
> > Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
>
> > That's a bit different. I thought you were talking about
> > processors with 100M elements. Color me much less impressed.
> > Surely we're not talking about standard type RAM here, or are we?
>
> > This doesn't put it in quite the same ballpark as the brain; the
> > machine has as many logic elements as a brain has neurons
> > (actually 100G is a lower estimate) but a neuron has considerably
> > more functionality than a simple logic element and the neuron does
> > have that big fan-out.
>
>now we are back to the starting point. A neuron is _not_ equivalent
>to a CPU. You suggested I consider 'logic elements' as the
>equivalent. I cited numbers in terms of logic elements.
No, I suggested that a neuron would be equivalent to a number of logic
elements, using 10 as a WAG. That's probably low and it depends on the
neuron. A neuron can be considered as a digital "element" in that it
fires or it doesn't. It isn't a case of something must be a logic
element or it must be a CPU.
Your claim was that the biggest machines were comparable to the brain.
It looks to me as though we still have a ways to go.
>the point is that we can't do the comparison, because, in the sense
>that a computer does, a brain doesn't 'compute'.
We disagree. I say that both compute but they use different
architectures and different kinds of components.
Nick Merritt
> In article <36b621aa...@nntp.ix.netcom.com>, mat...@ix.netcom.com
(Matt Silberstein) writes:
[snip]
> 6. Computer simulations are, by definition, mathematical models
> of the things being simulated. The models abstract away
> from the real thing. They deal only with some mathematical
> aspects of the thing being modelled.
> They are not the thing itself.
Look, let's assume evolution by natural selection *is* an algorithm (and
is not just being *described by* an algorithm).
Which leads to a question which goes to the base of these matters: is a
virtual organism which evolves by this algorithm a simulation or not?
Personally I don't think so. You are not just simulating the evolutionary
algorithm, you are actually running it in silico.
>
> For example, a simulation of a nuclear power-plant
> may model the nuclear reactions and the energy transfer,
> but the simulated energy cannot power a real light-bulb.
The key word here is 'model'. The algorithm we write to model these
processes are almost certainly not the processes operating in nature. I am
arguing that the algorithms used in the likes of Tierra *are*
fundamentally (one of) the processes operating in nature.
> Nor can a simulation of an aircraft acutally tranport people
> or things. ``the map is not the territory''
Would an in-silico thinking being be a simulation or a thinking being? Or
would you agree with the results of a Turing test and decide such
distinctions may in some cases just not be useful?
> Matt's basic thesis is that, in the case of life,
> the map *is* the territory. This seems to me, and to Marty,
> to be a fundamental semantic error.
> What is so special about life that, unliek all other
> simluations, the map really *is* the territory?
[snip]
> We're all adult enough to see who is being intellecutally dishonest
> here, and who isn't.
I don't think your aggressive tone is helping your argument. There's a
flourishing field of a-life studies which is doing very well at the moment
by not worrying about semantics. Let's argue about the validity of their
results in 20 years when they've had the chance to either produce
something amazing, or fail.
Nick
Nick Merritt
<usene...@usenet.nospam.fogey.com> wrote:
> The Emperor's New Mind
> Shadows of the Mind.
>
> There is a third, of which he is one of the editors, called _The
> Large, the Small and the Human Mind_, which I have not read.
>
> As to his arguments: I think he is, in part, correct, that is, that
> his conjecture that the brain relies on physical phenomena not yet
> discovered that can not be modeled by a computable function, is
> true.
Well it seems to me that only mathematicians reckon Penrose has anything
going for him on this point. I something wonder if it's more because
mathematicians have a bias about the 'specialness' of their discipline
than other people. Only in mathematics, it seems, can someone of Penrose's
stature get away with philosophising about its Platonic nature.
There have been strong counter-points made about the Godel problem. They
mainly revolve around the brain's ability to evolve solutions to novel
problems on the fly (something Penrose say little about in either of his
two books.)
Nor is Penrose even original. I remember reading this point about Godel
and non-computability on a book on psychology published in the 1970s - the
author was dismissive then too.
> I believe that Penrose has uncovered a significant part of the reason
> that AI has made so little progress in 40 years by comparison to other
> disciplines in computer science.
The reason so little progress has been made in AI has had more to do with
notions that you can actually design a complex intelligent entity from the
top down. This is a paradigm that has been well and truly shattered in the
last ten years. (Of course, this lack of progress has also had a lot to do
with the difficulty of the business of the brain and even there great
strides have been made in the last five years).
[snip]
Nick
PZ Myers
wrote:
>Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
>
>>>>>>> Richard Harter pounded silicon into:
>>
>> > Marty Fouts <usene...@usenet.nospam.fogey.com> wrote:
>>
>
>> > That's a bit different. I thought you were talking about
>> > processors with 100M elements. Color me much less impressed.
>> > Surely we're not talking about standard type RAM here, or are we?
>>
>> > This doesn't put it in quite the same ballpark as the brain; the
>> > machine has as many logic elements as a brain has neurons
>> > (actually 100G is a lower estimate) but a neuron has considerably
>> > more functionality than a simple logic element and the neuron does
>> > have that big fan-out.
>>
>>now we are back to the starting point. A neuron is _not_ equivalent
>>to a CPU. You suggested I consider 'logic elements' as the
>>equivalent. I cited numbers in terms of logic elements.
>
>No, I suggested that a neuron would be equivalent to a number of logic
>elements, using 10 as a WAG. That's probably low and it depends on the
>neuron. A neuron can be considered as a digital "element" in that it
>fires or it doesn't. It isn't a case of something must be a logic
>element or it must be a CPU.
Although, just to complicate matters a bit more, the significant signal
in a neuron's output may be *rate* of fire...so it isn't really digital
at all.
>
>Your claim was that the biggest machines were comparable to the brain.
>It looks to me as though we still have a ways to go.
>
>>the point is that we can't do the comparison, because, in the sense
>>that a computer does, a brain doesn't 'compute'.
>
>We disagree. I say that both compute but they use different
>architectures and different kinds of components.
I'm just curious...what is the definition of 'computer' that excludes
the brain?
--
PZ Myers
Clark Dorman
I believe that an additional complicating factor is that a significant
neuronal output may be its synchronization, or lack thereof, with
other neuronal firings. I remember going to a talk where someone was
proposing the synchronization as a possible solution to a binding
problem.
The binding problem in question is the following: when you hear or
read the sentence "John gave Mary the book", the objects John, Mary,
and book are represented in the brain (this is a generalization of the
'grandmother cell' idea only the representation is a pattern of active
neurons rather than one or a single cluster). The verb to give is
also represented somehow. The question is how does the brain bind the
giver (John) with the concept of giver, and the receiver (Mary) with
the concept of recevier, and the object given (book) with the concept
of the object given. Another way to pose the problem is, How does
the brain differentiate, through binding subjects, verbs, and
predicates, between "John gave Mary the book" and "Mary gave John the
book", or even "The book gave John to Mary"?
The speaker in question published his ideas on synchronization as a
solution to this problem in Brain and Behavioral Sciences (BBS). It
was a 'lively' talk.
Even if his idea was wrong (and I think it probably is or at least a
gross simplification), there is synchronization of neuronal firings
within the brain (hence waves) and some neurons fire with the waves,
some don't, and some sometimes do and othertimes don't. I'm not
convinced that the sychronization is not a separate factor.
--
Clark
Richard Harter
Will you accept quasi-digital?
>>Your claim was that the biggest machines were comparable to the brain.
>>It looks to me as though we still have a ways to go.
>>
>>>the point is that we can't do the comparison, because, in the sense
>>>that a computer does, a brain doesn't 'compute'.
>>
>>We disagree. I say that both compute but they use different
>>architectures and different kinds of components.
>
>I'm just curious...what is the definition of 'computer' that excludes
>the brain?
Computing is something that is done by machines that Marty is accustomed
to working with.
PZ Myers
wrote:
No.
Is that a good digital answer?
As Clark also just pointed out, phase or coincidence in a battery of
neurons may also be important. I've been working with a pool of neurons
that seem to just spew transmitter in the general direction of their target,
so we've also got graded responses due to diffusion. Then there's all the
neuromodulatory stuff, and field effects...there are just too many more options
than "off" or "on" to allow the simple digital answer.
>
>>>Your claim was that the biggest machines were comparable to the brain.
>>>It looks to me as though we still have a ways to go.
>>>
>>>>the point is that we can't do the comparison, because, in the sense
>>>>that a computer does, a brain doesn't 'compute'.
>>>
>>>We disagree. I say that both compute but they use different
>>>architectures and different kinds of components.
>>
>>I'm just curious...what is the definition of 'computer' that excludes
>>the brain?
>
>Computing is something that is done by machines that Marty is accustomed
>to working with.
Are you suggesting that he doesn't work with his brain? Or am I, by force
of long habit, reading far too much subtlety into your comment?
--
PZ Myers
Jonathan Stone
|> >Um, before we go getting all judgemental, let us try to remember that
|> >digital computers serve a different purpose than human brains. What
|> >the designers of computer systems do isn't 'wrong', it just has to
|> >meet different requirements.
|>
|> Yes, exactly. I wasn't trying to be judgemental.
Why the insult to computer architects, then? :->
|> >Agreed. There's somebody at MIT whose name I can never recall who took
|> >this observation to heart and has a whole different approach to robot
|> >building that can be viewed, in AI terms, as 'bottom up' rather than
|> >the usual 'bottom down' approach of the AI community.
|>
|> Yes, I've seen those.
Ha. Rodney Brooks. (i wrote a paragraph about brooks yesterday and
edited it out). The problem many researchers see with Brooks is that
he's never been able to articulate what his ``subsumption
architecture'' is in a way that lets others apply it successfully.
Many people in the CS community is skeptical of Brooks' research
programme, basically for that reason. Do you *really* get
`intelligence' out of stacking lots of smalll, specialized systems on
top of each other? Is there *really* a methodolgy behind it, or is it
just an accumulation of hacks?
Its a complex argument, too much to do justice to here.
|> > > Does anyone believe any computer could simulate the complexity
|> > > involved in sorting out multiple sensory modalities to navigate
|> > > and operate a complex flying machine to a piece of rotting fruit,
|> > > where it will land, eat, groom, seek out members of the opposite
|> > > sex of the same species, go through elaborate mating rituals,
|> > > mate, and reproduce?
|> >
|> >marvin minsky does. so do most ai researchers, and, apparently now,
|> >the artificial life fadists.
|> >
|> > > I don't think so.
|>
|> >me either.
Why on earth not? Do you think brains -- even fly brains -- are
non-computable? Are you dualists? Do you think the problem is
computationally intractable? Or is it somenthing else?
|> Although I would qualify it a bit to say they can't do it now, or in the
|> immediately foreseeable future. But it ought to be possible. Of course, it
|> would mean abandoning many of the pragmatic goals of computer design to
|> make something that maybe can't add two numbers together very quickly,
|> but does something else without immediate applicability quite well.
No, Paul. It would *not* mean abandoning computer architecture.
What you're sayin violates Church's thesis. Which says, in essence, If
you compute do it by ``adding lots of numbers together'', then you can
do it with *any* turing-complete formalism. So, if computers can do it
at all, *any* Turing-complete device can do it.
If you're saying the problem really needs a new way of thinking about
and describing the problem, fine. All that means is that we need a new
programming language or methodology to make progress.
But the choice of which hardware computer architecture to use is
irrelevant -- unless you start adding in other criteria, like cost,
or size. Or unless you disprove the Church-Turing thesis.
Unless you do that, your're in the same territory as
Einstein-was-wrong, faster-than-light kooks.
And if you *do* have an answer, all I'd ask is an invitation to the
Fields medal and Turing award presentations ;-).
Jonathan Stone
|> In article <793b32$fl1$9...@nntp.Stanford.EDU>, jona...@DSG.Stanford.EDU wrote:
|>
|> > In article <36b621aa...@nntp.ix.netcom.com>, mat...@ix.netcom.com
|> (Matt Silberstein) writes:
|>
|> [snip]
|>
|> > 6. Computer simulations are, by definition, mathematical models
|> > of the things being simulated. The models abstract away
|> > from the real thing. They deal only with some mathematical
|> > aspects of the thing being modelled.
|> > They are not the thing itself.
|>
|> Look, let's assume evolution by natural selection *is* an algorithm (and
|> is not just being *described by* an algorithm).
|>
|> Which leads to a question which goes to the base of these matters: is a
|> virtual organism which evolves by this algorithm a simulation or not?
|>
Thats the question *you* happen to see. I assert that begs the
question: how complete is your model? I think you even refute it
below, yourself.
|> Personally I don't think so. You are not just simulating the evolutionary
|> algorithm, you are actually running it in silico.
Fine. Suppose we have an algorithm which models biological evolution.
(and the key word is *model*: thats what science *does*). Suppose we
run it in silicon. It's *still* a model.
Precisely as if we code up the entire Standard Model of high-energy
physics and simulate some bunch of particles. Are we running a
simulation, or have we created a new universe in silico?
That's *exactly* the same question as whether the model of biological
evolution is *really* biolgocal evolution. How accurate is the model?
How do we know it captures all of what's going on?
Do you really not see that?
Our regular from Toronto might argue vociferously, for example, that
your models dont accurately capture genetic drift. (I'm not saying
they do or dont, because we havent got to specific models yet).
|> > For example, a simulation of a nuclear power-plant
|> > may model the nuclear reactions and the energy transfer,
|> > but the simulated energy cannot power a real light-bulb.
|>
|> The key word here is 'model'. The algorithm we write to model these
|> processes are almost certainly not the processes operating in nature. I am
|> arguing that the algorithms used in the likes of Tierra *are*
|> fundamentally (one of) the processes operating in nature.
So what? Physicissts run simluations too. The algorithms used in
particle simulations *are*, algorithmically, fundmendmtally ``one of''
the processes operating in nature. That doesnt make the simulations
real physical systems. Why is `life' any different?
There seems room to argue that calling it ``alive'', rather than a
simulation of life, is intellecutal fraud. Except that the strong
proponents really do see it that way.
|> > Nor can a simulation of an aircraft acutally tranport people
|> > or things. ``the map is not the territory''
|>
|> Would an in-silico thinking being be a simulation or a thinking being? Or
|> would you agree with the results of a Turing test and decide such
|> distinctions may in some cases just not be useful?
No, they'd have created an artifical intelligence that passses some
form of the turing test. Whether its ``real'' intelligence or a
simulation is left open. That's the difference.
|> > Matt's basic thesis is that, in the case of life,
|> > the map *is* the territory. This seems to me, and to Marty,
|> > to be a fundamental semantic error.
|>
|> > What is so special about life that, unliek all other
|> > simluations, the map really *is* the territory?
|>
|> [snip]
|>
|> > We're all adult enough to see who is being intellecutally dishonest
|> > here, and who isn't.
|>
|> I don't think your aggressive tone is helping your argument.
Probably not, but it gets very, very wearing being polite to opponents
who are a) hypocritical and b) dishonest when you point out basic
facutal flaws in their understanding. Go look at Dejanews.
But when people start making arguments that are *already* disproven,
mathematically and rigorously, then its fair game to point it out to
them. When they fail to understand *why* their pet ideas violate
fundamental results Turing-equivalence or Church' thesis, and you tell
them that, and they ignore it, then an ``agressive tone'' is, perhaps,
justified, wouldn't you say?
|> There's a
|> flourishing field of a-life studies which is doing very well at the moment
|> by not worrying about semantics.
Fine. If they do good science, more power to them. Sincerely. But
that doesnt mean we shouldn't worry about semantics. How else can we
interpret what their results and their conclusions about them really
mean?
|> Let's argue about the validity of their
|> results in 20 years when they've had the chance to either produce
|> something amazing, or fail.
No. The argument we're having here is whether simluations of life is
*real* life, not just `artifical' life. As you observe, that's
necessarily an argument about semantics.
And the Santa Fe institute can get valid, scientiifc results
whichever way the argument about sematics goes.
Jonathan Stone
|> Is that a good digital answer?
|>
|> As Clark also just pointed out, phase or coincidence in a battery of
|> neurons may also be important. I've been working with a pool of neurons
|> that seem to just spew transmitter in the general direction of their target,
|> so we've also got graded responses due to diffusion. Then there's all the
|> neuromodulatory stuff, and field effects...there are just too many more options
|> than "off" or "on" to allow the simple digital answer.
Paul, what does that have to do with anything?
Computers are quite adept at modelling phenomomena which, for all
intents and purposes, are continuous. Are you familiar with Nyquist's
theorem? All that's required is a simulatoin that sample the state at
twice the peak firing rate.
If only we knew how it was wired up and all what it all meant ;).
[snip]
|> >>I'm just curious...what is the definition of 'computer' that excludes
|> >>the brain?
|> >Computing is something that is done by machines that Marty is accustomed
|> >to working with.
|>
|> Are you suggesting that he doesn't work with his brain? Or am I, by force
|> of long habit, reading far too much subtlety into your comment?
If you're reading it, its probably there.
PZ Myers
>In article <myers-01029...@bio-101.bio.temple.edu>,
my...@astro.ocis.temple.edu (PZ Myers) writes:
>
>|> >Um, before we go getting all judgemental, let us try to remember that
>|> >digital computers serve a different purpose than human brains. What
>|> >the designers of computer systems do isn't 'wrong', it just has to
>|> >meet different requirements.
>|>
>|> Yes, exactly. I wasn't trying to be judgemental.
>
>Why the insult to computer architects, then? :->
It wasn't. Biology and computers have different goals -- so trying
to pigeonhole one into the other is "wrong".
>
>
>|> >Agreed. There's somebody at MIT whose name I can never recall who took
>|> >this observation to heart and has a whole different approach to robot
>|> >building that can be viewed, in AI terms, as 'bottom up' rather than
>|> >the usual 'bottom down' approach of the AI community.
>|>
>|> Yes, I've seen those.
>
>Ha. Rodney Brooks. (i wrote a paragraph about brooks yesterday and
>edited it out). The problem many researchers see with Brooks is that
>he's never been able to articulate what his ``subsumption
>architecture'' is in a way that lets others apply it successfully.
>Many people in the CS community is skeptical of Brooks' research
>programme, basically for that reason. Do you *really* get
>`intelligence' out of stacking lots of smalll, specialized systems on
>top of each other? Is there *really* a methodolgy behind it, or is it
>just an accumulation of hacks?
>
>Its a complex argument, too much to do justice to here.
And I'm not pretending to be qualified to know what approach works
better than another.
>
>|> > > Does anyone believe any computer could simulate the complexity
>|> > > involved in sorting out multiple sensory modalities to navigate
>|> > > and operate a complex flying machine to a piece of rotting fruit,
>|> > > where it will land, eat, groom, seek out members of the opposite
>|> > > sex of the same species, go through elaborate mating rituals,
>|> > > mate, and reproduce?
>|> >
>|> >marvin minsky does. so do most ai researchers, and, apparently now,
>|> >the artificial life fadists.
>|> >
>|> > > I don't think so.
>|>
>|> >me either.
>
>Why on earth not? Do you think brains -- even fly brains -- are
>non-computable? Are you dualists? Do you think the problem is
>computationally intractable? Or is it somenthing else?
None of the above. I think it is computable (I don't think the brain
is anything but a computer), and I'm certainly not a dualist. I should
have said "any computer NOW". I don't think the hardware or the software
is anywhere near the level that we could emulate a fly, YET.
>
>
>|> Although I would qualify it a bit to say they can't do it now, or in the
>|> immediately foreseeable future. But it ought to be possible. Of course, it
>|> would mean abandoning many of the pragmatic goals of computer design to
>|> make something that maybe can't add two numbers together very quickly,
>|> but does something else without immediate applicability quite well.
>
>No, Paul. It would *not* mean abandoning computer architecture.
>What you're sayin violates Church's thesis. Which says, in essence, If
>you compute do it by ``adding lots of numbers together'', then you can
>do it with *any* turing-complete formalism. So, if computers can do it
>at all, *any* Turing-complete device can do it.
>
>If you're saying the problem really needs a new way of thinking about
>and describing the problem, fine. All that means is that we need a new
>programming language or methodology to make progress.
>
>But the choice of which hardware computer architecture to use is
>irrelevant -- unless you start adding in other criteria, like cost,
>or size. Or unless you disprove the Church-Turing thesis.
>Unless you do that, your're in the same territory as
>Einstein-was-wrong, faster-than-light kooks.
>
>And if you *do* have an answer, all I'd ask is an invitation to the
>Fields medal and Turing award presentations ;-).
Ah, now I DID qualify my comments above: we "can't do it now, or in the
immediately foreseeable future". I'm not talking about a mathematical
entity, a hypothetical Turing machine that could simulate every last
capability of a fly. I believe (without any deep grounding in computer
science) that that is possible. I will even believe the formal possibility
that someone could write a program in Apple BASIC running on a 1-mHz 6502
that executes a simulator off of a paper tape that could do everything a
fly can compute...at glacial speed.
If you want to do something like that in our lifetimes and at a speed comparable
to what a fly can do, I think you would have to abandon current ideas about
computer architecture, though. That's not denigrating the current state of
the art in computers, it's pointing out that some computer scientists seem to
lack an appreciation for the sophistication underlying even the simple
nervous systems around here. And it seems to me that the direction computers
are going is not in the same direction that fly brains or mouse brains or
human brains have gone, so I don't see that human-like computation is
achievable or even desired imminently.
--
PZ Myers
Jonathan Stone
|> >>>>> PZ Myers pounded silicon into:
|>
|>
|>
|> > Although I would qualify it a bit to say they can't do it now, or
|> > in the immediately foreseeable future. But it ought to be
|> > possible. Of course, it would mean abandoning many of the
|> > pragmatic goals of computer design to make something that maybe
|> > can't add two numbers together very quickly, but does something
|> > else without immediate applicability quite well.
|>
|> design'. There's also the constraints of computer construction.
Please note that ` constraints of computer construction' means
something like ``hard physical constraints of running wires around
sub-micron features in three dimensions''.
Its relatively easy in 2-d, where the features and wires are painted
onto a crystal substrate with UV or X-ray lithography and deposited
via chemical processes. In 3-d, its an unsolved problem.
|> would be relatively trivial to design a 'neuron-model' in very dense
|> silicon, but there's no way to wire up all the neural models to each
|> other in 3d the way the brain is.
I've pointed that out too. I'm not sure the point is getting across,
though.
|> unless we learn to literally 'grow' such a structure, we're not going
|> to be able to engineer it simply because of the wiring problem.
Um. But if Richard Harter (and PZ Myers?) is right, then it really
doesnt matter: if the brain really is a computer then, from
Turing-equivalence, we can simulate it faithfully on existing
architectures.
|> But beyond that, IMO, Penrose is right and consciousness is not
|> computable.
I've never been able to make head or tail of Penrose' argument.
Jonathan Stone
|> In article <79567b$88h$1...@nntp.Stanford.EDU>, jona...@DSG.Stanford.EDU wrote:
|>
|> >In article <myers-01029...@bio-101.bio.temple.edu>,
|> my...@astro.ocis.temple.edu (PZ Myers) writes:
|> >
|> >|> >Um, before we go getting all judgemental, let us try to remember that
|> >|> >digital computers serve a different purpose than human brains. What
|> >|> >the designers of computer systems do isn't 'wrong', it just has to
|> >|> >meet different requirements.
|> >|>
|> >|> Yes, exactly. I wasn't trying to be judgemental.
|> >
|> >Why the insult to computer architects, then? :->
|>
|> It wasn't. Biology and computers have different goals -- so trying
|> to pigeonhole one into the other is "wrong".
Well, okay, so it wasn't intended. But saying that computer architects
were doing it all wrong--``putting the pieces together all wrong''
does seem like a pretty harsh insult. Heck, even to me: i've worked on
papers published in architecture conferences. q(NB: Unlike almost all
other fields, in CS, the top conferences are single-track, fully
refereed and with lower acceptance rates than most journals.)
|> And I'm not pretending to be qualified to know what approach works
|> better than another.
Me neither, but I think I'm at least qualified to have an opinion.
[ snip why cant computers simulate a fly brain ]
|> >Why on earth not? Do you think brains -- even fly brains -- are
|> >non-computable? Are you dualists? Do you think the problem is
|> >computationally intractable? Or is it somenthing else?
|>
|> None of the above. I think it is computable (I don't think the brain
|> is anything but a computer), and I'm certainly not a dualist. I should
|> have said "any computer NOW". I don't think the hardware or the software
|> is anywhere near the level that we could emulate a fly, YET.
Fine. Thats a technological argument. I'm happy to agree with that.
Perhaps its not a big difference to you, but to an active resarcher
in computer sciece, that's all the difference in the world. I suppose
is' a bit like terminology differences that Richard and Marty Fouts
have been tripping over.
Its like the people who are saying that ANNs or cellular automata can
or will do things that `conventional' computing cannot. Those people
are talking solid waste.
But just personally, I'd say many of the control problems you've
talked about can be handled, one by one, quite readily by current
hardware. If ten-year-old computers can fly a space shuttle, or model
the aerodynamics of a bumblebee, they should be able to do
flight-control for a fly. And there are quite simple and effective
models of (for example) how bees navigate while flying into flowers.
The part we really dont understand is how to put all those pieces
together into a sensible whole, and thats where Rodney Brooks is
watched with interests from both pro and con.
[snip]
|> >And if you *do* have an answer, all I'd ask is an invitation to the
|> >Fields medal and Turing award presentations ;-).
|>
|> Ah, now I DID qualify my comments above: we "can't do it now, or in the
|> immediately foreseeable future". I'm not talking about a mathematical
|> entity, a hypothetical Turing machine that could simulate every last
|> capability of a fly. I believe (without any deep grounding in computer
|> science) that that is possible. I will even believe the formal possibility
|> that someone could write a program in Apple BASIC running on a 1-mHz 6502
|> that executes a simulator off of a paper tape that could do everything a
|> fly can compute...at glacial speed.
Fine. We're talking the same language.
*However*, the thing that got this argument really going was when
somebody said (to paraprhase) that an Unobtanium-II running the exact
same algorithm at lightning speed would be ``alive'', by virtue of
parallels between 'real' life and the objects and processes within the
simulation. But at the same time, your 6502 using a paper-tape as a
Turing store to execute the identical simulation would *not* be alive.
In CS, that *is* a seqeunce-heme level blunder. No, its worse.
|> If you want to do something like that in our lifetimes and at a speed comparable
|> to what a fly can do, I think you would have to abandon current ideas about
|> computer architecture, though. That's not denigrating the current state of
|> the art in computers, it's pointing out that some computer scientists seem to
|> lack an appreciation for the sophistication underlying even the simple
|> nervous systems around here.
|> And it seems to me that the direction computers
|> are going is not in the same direction that fly brains or mouse brains or
|> human brains have gone, so I don't see that human-like computation is
|> achievable or even desired imminently.
Oh, it's desired all right. The first fundamental problem is we dont
know enough physics to make computers (in the broad sense) that have
three-dimensional interconnections. The second fundamental problem is
that even if we *did*, we have no idea at all how to program them into
something like a human brain.
If we did have a cule, we could program it up and then emulate it,
even at at glacial speed, on current computers. People have tried all
sorts of approaches, and none of them (including ANNs) work very well
at capturing the spectrum of human intelligence.
I'm kinda curious, though. If you're that negative about mathematical
simulations capturing the behaviour of brains, what's your stance on
artificial life? What does it mean to have an artifical cricket if
there's no way, even in the foreseeable future, to do an acceptable
job of simulating a Drosophila melanogaster brain?
PZ Myers
>In article <myers-01029...@bio-101.bio.temple.edu>,
my...@astro.ocis.temple.edu (PZ Myers) writes:
>|> In article <36b61339....@199.0.65.59>, c...@tiac.net (Richard Harter)
>|> wrote:
>
>|> Is that a good digital answer?
>|>
>|> As Clark also just pointed out, phase or coincidence in a battery of
>|> neurons may also be important. I've been working with a pool of neurons
>|> that seem to just spew transmitter in the general direction of their
target,
>|> so we've also got graded responses due to diffusion. Then there's all the
>|> neuromodulatory stuff, and field effects...there are just too many
more options
>|> than "off" or "on" to allow the simple digital answer.
>
>Paul, what does that have to do with anything?
>
>Computers are quite adept at modelling phenomomena which, for all
>intents and purposes, are continuous. Are you familiar with Nyquist's
>theorem? All that's required is a simulatoin that sample the state at
>twice the peak firing rate.
I might ask, what does that have to do with what I said?
The question is: should we think of neurons as digital devices? I say no,
and gave some reasons.
I was NOT claiming that neurons are analog devices, therefore digital
computers cannot simulate them. I have no problem with the idea that
digital computers can model analog devices.
I think you are trying a little too hard to find what I say to be in
opposition to your ideas. I don't have your background, and all I can
do when you talk about computer architecture is nod my head and figure
you know what you are talking about. My only input here is to point out
that computer scientists may be underestimating the capabilities of
the computers I do know a little bit about, biological brains, and that
from my perspective I don't see man-made computers converging in any way
on the properties of brains. That's not to deny the amazing potential
and dazzling brilliance of you guys -- you're just going in a different
direction.
--
PZ Myers
PZ Myers
[snip]
>I'm kinda curious, though. If you're that negative about mathematical
>simulations capturing the behaviour of brains, what's your stance on
>artificial life? What does it mean to have an artifical cricket if
>there's no way, even in the foreseeable future, to do an acceptable
>job of simulating a Drosophila melanogaster brain?
Whatever gave you the idea I'm negative about it? I just think it is
a very, very hard problem that isn't going to be easily solved in the
near future with off-the-shelf components and software written by a
grad student in Lisp. That's not negative, that's realistic.
I think a-life is neat. It's got a long ways to go, but that's not a
strike against it. I can't explain how a zebrafish develops yet, so should
I just give up? Would it be OK if I just tried to work on some small
piece of the story, and come up with a few provisional ideas?
--
PZ Myers
Jonathan Stone
|> >|> so we've also got graded responses due to diffusion. Then there's all the
|> >|> neuromodulatory stuff, and field effects...there are just too many
|> more options
|> >|> than "off" or "on" to allow the simple digital answer.
|> >
|> >Paul, what does that have to do with anything?
|> >
|> >Computers are quite adept at modelling phenomomena which, for all
|> >intents and purposes, are continuous. Are you familiar with Nyquist's
|> >theorem? All that's required is a simulatoin that sample the state at
|> >twice the peak firing rate.
|>
|> I might ask, what does that have to do with what I said?
|>
|> The question is: should we think of neurons as digital devices? I say no,
|> and gave some reasons.
|>
|> I was NOT claiming that neurons are analog devices, therefore digital
|> computers cannot simulate them. I have no problem with the idea that
|> digital computers can model analog devices.
|>
|> I think you are trying a little too hard to find what I say to be in
|> opposition to your ideas.
Nope, not at all. I can't make any sense of the distinction you're
trying to make. I'm sure theres one there, though.
|> I don't have your background, and all I can
|> do when you talk about computer architecture is nod my head and figure
|> you know what you are talking about.
Does that include my appeals to the Church-Turing thesis?
|> My only input here is to point out
|> that computer scientists may be underestimating the capabilities of
|> the computers I do know a little bit about, biological brains, and that
|> from my perspective I don't see man-made computers converging in any way
|> on the properties of brains.
|> That's not to deny the amazing potential
|> and dazzling brilliance of you guys -- you're just going in a different
|> direction.
Yes. But I am appealing to a famous result[*] in the foundations of
mathematics and computer science, which says that if there exists at
least one kind of computer, of any architecture you choose, that can
do what biological brains do, then *any* turing-equivalent device can
do it.
Once we figure out how, it doesnt matter what approach we take,
something wiht an interconnect resembling neurons or osmething
resembling a von-neumann machine, or something in between.
The only real problem is figuring out how to *do* it.
Once we get that, then its all down to engineering trade-offs.
Thats not appealing to computer architecture, thats appealing to
mathematics. Mathematically, if you accept that biological brains are
computable, they are Turing-equivalent, then mathematics says it
doesnt *matter* if the hardware archicetures go in different
directions. If the brain is Turing-computable (which you say it is)
then (a) it can emulate a turing-machine, and (b) a turing machine can
emulate it.
Or, to put it another way, it matters no more than whether Visual
Basic or Java are diverging: its the exact same fundamental question.
Anything you can do in one, you can do in the other.
The standard theory-of-computation proof is to write a Java
interpreter in Visual Basic, and a Visual Basic interpeter in Java.
And when you say what the brain does is computable, then you're saying
it falls into the same class. (that includes, btw, the people who
think the brain is modellable by cellular automata, since they are
Turing-equivalent too.) All you'd have to do is write a Von-Neumann
interpreter for your ``brain'', and a ``brain'' intepreter for a
traditional von-neumann computer. Once you do that, then they are,
rigorously, mathematically, formally, equivalent.
The reason why computer architects dont try and emulate wetware
architecture is because of the physics of manufacturing small wires in
3-d. The rest is just down to the bst programming langauge for
a problem.
So I guess I really truly dont understand why you think CS people dont
appreciate biological brains. After all, passing restricted,
behaviourally-equivalence test with a human brain is *the* Grand
Challenge in CS.
[*] more formally, i guess its a conjecture, but a conjecture of the
kind which Godels theorem says that if its wrong, we'll never be able
to prove it. Probably. Richard Harter said it much more eloquently
than I can, but we're both fudging it a bit.
Jonathan Stone
|> In article <79587m$88h$3...@nntp.Stanford.EDU>, jona...@DSG.Stanford.EDU wrote:
[snip again]
wow. after that last humungous message, perhaps this is all
that needs to be said:
Just what is it about biological brains that you think CS folk are
underestimating? Or that people working in AI aren't looking at,
that they should be?
Jonathan Stone
|> >simulations capturing the behaviour of brains, what's your stance on
|> >artificial life? What does it mean to have an artifical cricket if
|> >there's no way, even in the foreseeable future, to do an acceptable
|> >job of simulating a Drosophila melanogaster brain?
|>
|> a very, very hard problem that isn't going to be easily solved in the
|> near future with off-the-shelf components and software written by a
|> grad student in Lisp. That's not negative, that's realistic.
I concur.
All I'm doing is obsevign that But Alan Turing says that if its doable
*at all*, its doable with off-the-shelf components, or even with
pencil-and-paper, if you have long enough to wait.
Some people here demonstrably havent understood that.
|> I think a-life is neat. It's got a long ways to go, but that's not a
|> strike against it. I can't explain how a zebrafish develops yet, so should
|> I just give up? Would it be OK if I just tried to work on some small
|> piece of the story, and come up with a few provisional ideas?
Fine. A-life is neat. Its a cool glitzy demo. They always impress.
And I've never said that simulation are not worth pursuing or that
they're not useful tools.
But anyone who loses track of the distinction between a scientific
theory (forget simulation models for awhile), and the real-world
things described by that theory, is not to be trusted. At least not as
a scientist.
And it seems to me thats thats exactly what the people who say a-life
really *is* life are doing. If it turns out they go back and verify
any _biological_ results they get from their simulation with real
life, fine, that's good science. But its precisely the need to
verifying the simulation predictions with real data which
proves that the simulations are *not* really alive: they're just
simulations, scientific tools.
OTOH, if the a-life people dont verify their conclusions against real
biological organisms, then they dont have any results in biology,
period. At least thats what *my* philosophy of science says.
I dont expect you to agree, but are we at least making progress on
understanding where the disagreement is?
Jonathan Stone
|> In talk.origins I read this message from
|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone):
|>
|> >In article <36b621aa...@nntp.ix.netcom.com>, mat...@ix.netcom.com (Matt Silberstein) writes:
|> >
|> >This is getting ridicuolus. Here is the list of questions again,
|> >for anyone who hasn't noticed that Matt is consistently avoiding
|> >answering them:
|> >
|> Even though I will, in general avoid all of Jonathan's posts in
|> future, this has enough content to ad hominem ration to be worth
|> dealing with.
|> >
|> > 1. Matt's definitions of `life' are metaphorical.
|> > THe interpretation of `metabolize' and `reproduce' is so loose
|> > that, as i see it, it includes flames and crystals.
|> >
|> > Are flames and crystals alive?
|>
|> I have answered this. Flames and crystals do have some of the
|> properties of life, but not much. On a scales they are quite close to
|> a non-crystal rock. I don't see life as a binary condition so I can't
|> answer yes/no.
But why, then, do you not acknowledge that you are changing the
accepted definition of `life'? As far as I'm aware, that has
historically been a binary condition. The famous problematic cases
are problems precisely because they don't fit that binary thresh-hold.
|> Now maybe
|> >
|> > 2. Living things do not have on/off switches or reset buttons.
|> > They cannot be paused (e.g, by powerfail or checkpoint
|> > mechanisms) and then restarted.
|> > Computer simluations can. How, then, are computer simulations
|> > alive?
|> >
|> I have answered this several times as well. I consider the current
|> ability to turn off and on ALife a significant drawback. It shows that
|> the current candidates are not very alive. But I point out that
|> viruses can be turned off and on as well. I will repeat. Life seems
|> like a continuum.
It seems that way to you. Possibly to many other people. but the
problem is that you are allowing absoutely everything-- physical,
chemical, biological, social, legal, computational -- onto that
spectrum.
Now, if you can come up with a coherent, consistent protocol for
ranking *EVERYTHING* on your continuum of life, something that
skeptics like me can apply, then you have come up with a well-defined
concept. So far you have not done that.
As Richard Harter has observed, the onus is on you to do that.
But even that wouldn't establish that you are talking about the conventional
concept of `life'. If you're claiming that, the onus is on you to
show that, too.
|> We have rocks on one end, cats on the other, and
|> viruses in the middle. Crystals and flames are close to rocks. Current
|> ALife is somewhere between a rock/crystal/flame and a virus.
This is not a useful protocol. *I* see crystals and flames as more
alive than IBM. IBM has the very significnat difference that its
components are not in phsyical contact with each other.
That's a key difference with the existing definitions of life. It's
why individual organisms are alive, yet colonies or eusocial creatures
or herds of individual organims are not alive.
|> > 3. Computer simulations do not acutally metabolize.
|> > The level of abstraction wehre we find similarties between
|> > a living thing and a computer simluation of a living
|> > thing does not actually exchange energy: instead, it
|> > exchanges information. (Any energy exchagne takes place
|> > purely within the computational devices themselves).
|> >
|> > Precisely how does exchange of information constitue
|> > *real* `metabolism' as opposed to a *smulation* of metabolism?
|> >
|> I am not sure why metabolism must necessarily mean energy (in a
|> physics sense) as opposed to meaning some sort of subsistence.
Because thats what the word ``metabolism'' *means*?
|> You
|> want to dismiss this as "metaphorical", go ahead. I see it as
|> exploring the ideas.
Fine. Go ahead and explore.
But if to change the meaning of words in that fashion, and then to say
that they still denote the same concept, is, in general, clearly not
true. (Unless you are a Po-Mo). If you change the interpretation of
`metabolism' and then argue that you a are talking about the same
concept as the *normal8 interpretation of biochemical metabolism, then
the onus is on *you* to prove that they are, indeed, the same concept.
If you want to say that IBM is alive or that simluations are alive,
then you have to establish that the sense in which you use `metabolize'
and `reproduce' is sufficiently close that we should decide it ist
the same thign.
This, you have not done. You have merely asserted it. If you
want to be taken seriously then the onus is on you to *show* it.
|> > 4. Computer simluations are merely comptuational models.
|> > They can be executed on a huge range of computational devices,
|> > all the way from supercomputers to pencil and paper.
|> >
|> > Mathematically, these computations are all identical:
|> > that is the Church-Turing thesis.
|> >
|> > Is a computation done on pencil and paper alive?
|> >
|> > If not, then Matt's orignal claim is refuted; there
|> > is something to life *APART* from the complexity and
|> > a homomorphism between biological organisms and simluations
|> > of biological organisms.
|> Rate of action *in the real world* is important.
In the *real* world, yes, But not in mathematical models such as
computer simluations. In simulations it is *utterly* irrelevant. That
is axiomatic. Please familiarise yourself with Turing equivalence. I
suggest the textbook by Papadimitrou. I can get a full bibilographic
reference if you require it.
If you say otherwise I'm afraid i will have to insist on full
bibliographic references to the peer-reviewed literature.
|> But no tIf there was such a
|> situation with a person doing the work on paper and pencil it might, I
|> suppose, be alive. But I doubt any of us could possible recognize it.
Good. What you said earlier was a ``sequence-heme'' level mistake.
I am pleased to see you have the sense to notice that and retract it.
But making it in the first place gives good grounds to be particularly
suspicious of your personal opinions: you are clearly not an expert or
even well-informed in the field.
that is *not* an ad-hominem, its just a statment of fact.
|> A paper and pencil implementation of a 3-D graphics program may be
|> formally equivalent to the program running on a computer, but I doubt
|> a person could recognize the resultant output as the same.
No, Matt. it *is* formally equivalent. Please familliarise yourself
with what Turing-equivalence means. Otherwise you are just Making it
clear that you are not au-fait with the relevant theory, and that is
not a good way to help your case. It just says we have to examine
everything *else* you say and look for similar misconceptions.
That, BTW, is part of what ``peer review'' means.
|> > 5. Which goedel numbers of computations are alive, and which
|> > are not, and why?
|> Why do I need to have some mechanism for determining this? So some may
|> and may not be. Maybe we will figure out some wonderful way to make
|> this determination, or maybe we won't. So what?
Do you sincerely not understand the point which this question is
probing at? I had thought it might elicit an embarassed retraction.
I'm sorry, I didnt realise you were that ignorant.
|> > 6. Computer simulations are, by definition, mathematical models
|> > of the things being simulated. The models abstract away
|> > from the real thing. They deal only with some mathematical
|> > aspects of the thing being modelled.
|> > They are not the thing itself.
|> >
|> > For example, a simulation of a nuclear power-plant
|> > may model the nuclear reactions and the energy transfer,
|> > but the simulated energy cannot power a real light-bulb.
|> > Nor can a simulation of an aircraft acutally tranport people
|> > or things. ``the map is not the territory''.
|> >
|> > Matt's basic thesis is that, in the case of life,
|> > the map *is* the territory. This seems to me, and to Marty,
|> > to be a fundamental semantic error.
|> >
|> You assert that they are simulations and you your label as an
|> argument.
No. I dont assert they are simulations. I observe the fact that they
*are* simuations. They are computations. That means they are
executions of a mathematical model. That much is purely definitional.
That' not an assertion, thats what computation *is*.
The particualr mathematical model is chosen because it's an abstraction
of certain things in the real world. That makes it a simulation.
This, too, is definitional.
|> The question is what are the important qualities of life.
But i dont agree at all that that is `the' question.
|> Reproduction is process, ISTM it does not matter if the system is
|> based on DNA or something else. And parts of ALife systems do, after a
|> fashion, reproduce. They don't simulate reproduction, they do it.
No. they dont ``do'' it, they *simulate* it. Please look at your own
words: ``after a fashion''. They do not *really* reproduce, any more
than simulations of physical particles *really* have momentum.
That is *exactly* the question we are arguing. And you are begging
the quesiton.
I am using well-accepted definitions from (a) mathematics and (b)
computer science. I'm using standard scientific epistemology as
applied to simulations.
If you want to claim some other defintions apply, fine. go ahead.
But once again, the onus of doing that is on *you*.
|> Now
|> maybe these systems do not have enough of the qualities of life and
|> maybe they never will. But that is an issue to discuss and explore,
|> not to assert.
But Matt, you are *not* discussing it. *You* are merely asserting it.
You assert that a-life is not merely simulation; you assert it is
*alive*. You want to make that claim, fine, go ahead. But the
onus is on you to substantiate it, not on me to refute it.
|> Arguing that tradition says one thing so we can't say
|> another does not fly.
No, Matt. If you want to change it, then the onus is on you
to to provide the (scientific) justification for it.
|> Nor does arguing that we have categorized the
|> systems are simulations so they are *only* simulations.
But simulations is exactly what they are. Rigorously and
mathematically. They are computations. If you want to show
they are more than that, then the onus is on you.
So far, all you have done is shown you lack sufficient
familiarity with theory of computation even to understand why
challeging questions from that field are even relevant.
|> > What is so special about life that, unliek all other
|> > simluations, the map really *is* the territory?
I cannot see anything at all which gives a substantive answer
to this question.
[snip]
|> >It's really up to Matt. He can either try and answer the questions
|> >substantively, or not.
|> >
|> >We're all adult enough to see who is being intellecutally dishonest
|> >here, and who isn't.
|>
|> Agreed.
Definitely.
Jonathan Stone
numbers are relevant to the question of whether computations are
alive? Or why `Which godel numbers are alive, and which aren't, and
why' is a pertinent question?
Ideally, An explanation would be aimed at an audience with limited
exposure to formal logic and theory of computation -- e.g., has read,
but not understood, Hofstadter's ``Godel, Escher, Bach''.
PZ Myers
What about those of us who have read GEB twice and *think* we understand
it? :-)
--
PZ Myers
Jonathan Stone
|> In article <795gh4$88h$1...@nntp.Stanford.EDU>, jona...@DSG.Stanford.EDU wrote:
|>
|> >Could someone -- Marty Fouts maybe -- give a quick run-down on Go\"del
|> >numbers are relevant to the question of whether computations are
[snip]
|> What about those of us who have read GEB twice and *think* we understand
|> it? :-)
Just for your, I'll try tomorrow, once I catch up with
thesis/paper-writing after yesterday's stint.
Stew Dean
Stone) wrote:
>In article <36b5804a...@news.dircon.co.uk>, stew...@webslave.dircon.co.uk (Stew Dean) writes:
>|> On 31 Jan 1999 20:54:44 -0500, jona...@Cup.DSG.Stanford.EDU (Jonathan
>|> Stone) wrote:
>
>
>|> Content = 0
>|>
>|> I think that proves Matt's point. Your lack of control does not help
>|> any points you do have (none in this case). As with any replies saying
>|> simply 'you're wrong' or using personal attacks it only helps confirm
>|> the point made by the person being attacked or told 'they're wrong
>|> because I said so' .
>|>
>|> I say this only in the hope we can keep these debates open and
>|> interesting.
>
>Um, no, I dont agree. the problem is that Matt really *doesnt* know
>what he's talking about.
And you totally failed to prove that in you last post! You made no
points. My point is you made no points in this or you last post. You
think Matt doesnt know what he's talking about does not make it so.
So far Matt is winning on points :)
Stewart Dean - ste...@webslave.dircon.co.uk
alife guide - http://www.webslave.dircon.co.uk/alife
Stew Dean
Stone) wrote:
>In article <36b5a22a...@199.0.65.59>, c...@tiac.net (Richard Harter) writes:
>|> jona...@Cup.DSG.Stanford.EDU (Jonathan Stone) wrote:
Oh - we're talking about complexity here...
>|> Stasis/complexity/chaos is the tip off.
>
>Sorry, to me that's just a tip-off for fuzzy new-age thinking, and I
>answered it accordingly. Celluare automata are just one more
>turing-equivalent formalism, and not a very interesting one at that.
So iterative systems don't interest you? Maybe that's the problem.
As for fuzzy new-age thinking - that tells us a lot about you
jonathan.