Re: Perceptual symbol systems

From: Traveler (traveler_at_nospam.com)
Date: 08/07/04 

Date: Sat, 07 Aug 2004 13:29:42 -0400

In article <4113f221$1_3@news.athenanews.com>, "Sergio Navega"
<snavega@intelliwise.com> wrote:

>"Traveler" <traveler@nospam.com> escreveu na mensagem
>news:v4g5h01i294h931ca0ji6ss9pnltg6j4hn@4ax.com...
>> In article <4112b4dc_6@news.athenanews.com>, "Sergio Navega"
>> <snavega@intelliwise.com> wrote:
>> >
>> >I may find some truth in the idea that intelligence involves the
>> >manipulation of "symbols", but not that kind of symbol that a
>> >philosopher (or a GOFAI researcher) thinks. The danger of using
>> >this word is to confuse "internal symbols" (patterns of activations
>> >which vaguely stand for something external) with "public symbols"
>> >(which are the physical markings that we use to communicate with one
>> >another).
>>
>> It is a good thing you saw fit to put "internal symbols" between
>> quotes because there are no such things. What I mean is that a neural
>> spike may be considered a symbolic event from the point of view of an
>> outside observer looking in, but from the point of view of the brain
>> itself, it is just another spike. So the mechanism of the brain could
>> not possibly be thought of as manipulating symbols.
>
>The big jump we have to give here is not to consider a single spike as
>representing one "symbol", but to look at the dynamical (and transient)
>behavior of a group of neurons as standing for some external stimuli
>(or at least standing for some specific properties of that external
>stimuli).

I agree that it does have correspondence with external phenomena, but
only from the outside observer's point of view. Not from the point of
view of the brain. In trying to understand brain function, it is
important, IMO, to put oneself in the shoes of an ontological
solipsist and ignore the external world. IOW, if you want to
understand the neuron, be the neuron! If you want to understand a cell
assembly, be the cell assembly! Neurons and assemblies do not see an
outside world. All they see are spikes and temporal patterns of
spikes. Nothing else. Spikes and patterns of spikes should be the
primary basis for a science of brain organization. The outside world
is important only as far as testing our models is concerned.

Certainly, we can use external observations as a guide which will
direct us toward possible models. But, in formulating our models, we
must not think in terms of symbols or knowledge representation. That
is a red herring and a waste of time.

>> >> What I am saying is that we must look for more fundamental principles.
>> >> Neurons do not manipulate symbols and they do not categorize.
>> >
>> >But a group of neurons (or the *behavior* of a group of neurons) may
>> >be equivalent (functionally equivalent) to some kind of symbol
>> >processing machine (symbol here understood in that "internal" sense
>> >I have mentioned above).
>>
>> Again, to an outside observer, yes. But not to the brain itself. This
>> should be clear to anyone willing to give it a modicum of thought.
>> Hence the reason that GOFAI and its supporters are bankrupt. This
>> realization should have changed the approach taken by the AI community
>> a long time ago. But I guess incestuous ideas begotten within a close
>> knit group only beget more incestuous ideas.
>
>I share your discredit toward GOFAI, but one have to ponder why
>they have believed that during so much time (and why there's still
>supporters of this idea).

It is a phenomenon that I have taken to calling "group think." It is a
behavior among groups wherein peer pressure and the willingness to
conform prevent members of the group from thinking outside the box.
This applies to all groups, including political, religious and
scientific ones. It is a form of self-imposed blindness.

> The fact is that sometimes we really act
>as if we were processing symbols.

We do process symbols. But that is an external behavior based on
external observation. It is not the basis of intelligence.

>Any solution conceived to be
>human-like will have to present similar behaviors (even if, internally,
>very different processes are running). Thus, I see GOFAI as
>presenting a series of constraints that our systems must obey.

If that is all that they did, I would have nothing to complain about.
But they went much further than that, to the detriment of AI research.
They declared that intelligence is based on symbol processing, which
could not be further from the truth, as I have shown.

>> >> Categorization is just one of many by-products of intelligence. I have
>> >> been preaching (for want of a better word) for a long time that
>> >> fundamental intelligence is temporal in nature.
>> >
>> >Perhaps I can also say that I've also been "preaching" that intelligence
>> >is a subset of the behavior of machines capable of pattern manipulation.
>> >Such pattern manipulation machinery uses, in our brain, temporally
>> >implemented processes.
>>
>> I disagree. To manipulate a symbolic pattern, one needs to know what
>> the pattern represents.
>
>I don't think so.

You are kidding me?

> But never mind, the kind of symbol I'm interested
>have some special properties.

This should not be dismissed so casually, IMO. The operational closure
of the brain is an important consideration.

>> After a symbol stand for something else,
>> otherwise it is not a symbol. The brain does not such thing. Why?
>> Because it has no idea what the pattern represents. A neural spike is
>> just a spike.
>
>One of the interesting things that have been developed in the last
>decade or so is the idea of "perceptual symbol systems", as proposed
>by Lawrence Barsalou. GOFAI fails because they use amodal symbols.
>In perceptual symbol systems one uses modal symbols, entities that
>stand for something (like any symbols must do) but that keep "something
>else" about that thing. In this regard, perceptual symbol systems
>conserve some qualities of the thing it represents.

Of course they do. But this is not what intelligence is based on.
Again, the intelligence does not see the outside world. It only sees
neural spikes and patterns of spikes. The intelligence may conclude
that there is an outside world but that is irrelevant to the working
of the neurons and cell assemblies. These do not care.

Try looking up the term "the operational closure of the nervous
system" on the web. Some psychologists are way ahead of the GOFAI
community. The excellent work of Alexander Riegler on anticipation
comes to mind:

http://www.univie.ac.at/constructivism/people/riegler/papers/riegler01anticipation.pdf

Louis Savain

Artificial Intelligence From the Bible:
http://users.adelphia.net/~lilavois/Seven/bible.html

Falsifiable Predictions:
http://users.adelphia.net/~lilavois/Seven/predictions.html

Relevant Pages

  • Re: Symbolic AI: Why Marvin Minsky and Curt Welch Are Out to Lunch
    ... is that spoken word not a symbol? ... connotation of "symbol" is a fixed material object), but spikes are ... able to duplicate the control functions performed by a human brain would ... What could possibly be more symbolic AI than "pulses are the discrete ...
    (comp.ai.philosophy)
  • Re: Perceptual symbol systems
    ... >>behavior of a group of neurons as standing for some external stimuli ... In trying to understand brain function, ... All they see are spikes and temporal patterns of ... Spikes and patterns of spikes should be the ...
    (sci.cognitive)
  • Re: 50 years later, Marvin Minsky still doesnt get it
    ... " But the human brain processes all information using spike signals." ... Every neuron in the brain produces spikes don't they? ... are discrete events as well. ...
    (comp.ai.philosophy)
  • Re: Medical image processing - need help pls
    ... I have a time series of pictures. ... And what I'm trying to find are spikes that occur in different brain ... pathologies. ...
    (sci.image.processing)