"It's got a good beat, and you can dance to it."

From: Glen Foy (spam33_at_butter.toast.net)
Date: 08/21/04 

Date: 21 Aug 2004 06:02:42 -0700

(This thread was posted in comp.ai.philosophy. I had intended to
cross-post it in sci.cognitive, but apparently forgot. Here it is for
those who do not read both groups.)

_________________
8/15/04

"It's got a good beat, and you can dance to it."

History doesn't record the name of the teenager who first gave ***
Clark that immortal summation, but the person was obviously a genius.

Music is remarkable, powerful, universally appealing, and it offers
profound clues as to the nature of intelligence. For AI research, it
is almost certainly a Rosette Stone.

A great deal of work has been done designing systems which can
generate music algorithmically. This is interesting hacking, but in
my view, it's putting the cart before the horse. What we should do
first is develop systems which can RESPOND to music rather than create
it. If the next neural network you design starts tapping its foot
while listening to the Indigo Girls belt out "Land of Canaan", you
will know that you are definitely on the right track ...

There are two essential issues here:

1) What does it mean to say that a computer model "responds" to
music?

and,

2) What does this "response to music" tell us about the fundamental
architecture of intelligence?

The answer to the first question is, in part, a result of the human
brain's "greedy parsing algorithm". Human intelligence tries to make
sense of the world as completely as possible. We sort, filter,
organize, conceptualize and generally massage raw perceptual data,
fitting it into our perceptual/conceptual schemes or creating new
schemes when necessary.

Music feeds the greedy parsing algorithm's need for COMPLETE AND
THOROUGH PROCESSING. Music is comprehensible in a way that everyday
reality is not. It's easy to parse, and it can be parsed more
completely, displaying an elaborate, consistent and varied internal
structure. Reading your monthly telephone bill just doesn't have the
same effect.

A computer model that "responds" to music would generate an internal
representation of the parse with an elaborate, consistent and varied
structure. It would produce a less interesting parse of your phone
bill. The difference would be quantifiable. Music is not unique in
this respect. A successful computer model would also respond to the
structure of a Shakespearean sonnet, a Cezanne still life or an
elegant mathematical proof.

But there is a sense in which music is unique. Music is rhythmic.
"It's got a good beat, and you can dance to it." The pulsed neural
networks of the human brain are also rhythmic. Some (possibly most)
of the information is stored in the spacing between spikes. Clearly,
music achieves much of its effect by being so closely tied to the
fundamental data structure of the hardware.

Here is a metaphor for you:

The brain is a nation of tribes. The tribes communicate with
drumbeats. While individual neurons only connect directly to
thousands of other neurons, they connect indirectly to billions. The
drumbeats of individual tribes propagate, and while they may be
modified, elaborated, attenuated, and transformed, you can be sure
that they echo in one form or another throughout the brain.

More on this, and especially question number 2, in a later post.

_________________
8/16/04

In the previous article I argued that the brain's processing of art is
unique, and that an understanding of art suggests a blueprint for the
fundamental architecture of intelligence. A computer model that
responds to art with an especially elaborate, consistent and varied
parse will be on the right track.

What this is saying, really, is that artists were the first
experimental cognitive scientists. Their interest was practical
rather than theoretical. They wanted to produce interesting music,
poetry, paintings, etc. Only rarely did they conceptualize. They
learned through trial and error and intuition what was effective at a
fairly low perceptual level, and they used these discoveries as the
building blocks of their artistic language.

Artist and musicians are fond of saying something "works" or it
doesn't "work". They mean this quite literally; it's not a metaphor.
An art object is a highly structured, interrelated and dynamic system.
 It achieves its effect by pressing all the right perceptual buttons.
These are the buttons that we need to understand and model. (For a
good description of the way the visual arts "work" see Rudolf
Arnheim's _Art and Visual Perception_ and _Visual Thinking_)

Now, it could be suggested that while art may teach us important
things about perception, it really doesn't have much to say about
abstract thought. I have argued previously in this forum that there
is probably not a sharp division between perception and conception.
Perception/conception is probably a layered structure. The very
highest levels of abstract thought could simply be seen as a
perception of the processes below. All thought may simply be
perception, and the algorithms used at the lower levels are probably
quite similar to the algorithms used at the higher levels. I won't
repeat these arguments here; if this interests you, the articles are
available at the website below. The point is that what we learn about
low-level perception from a study of art is probably directly
applicable to abstract thought as well.

>From a practical point of view, AI researchers are a long way from
developing a model that can appreciate the structure of a Bach fugue,
and I'm not seriously suggesting that that should be a goal, at least
not now. What I am suggesting is that we do some reverse engineering.
 We have these amazing artifacts called art objects. They were
created by the members of our species with the most acutely developed
perceptual abilities. Art objects achieve their intensity by
skillfully strumming the fundamental algorithms of perception. If we
can understand these algorithmic arpeggios, we will be very close, in
my opinion, to the heart of intelligence.

_________________
8/17/04

"An understanding of art suggests a blueprint for the fundamental
architecture of intelligence."

So what is this architecture of intelligence? Clearly it can't be
described in a 500 word newsgroup post, but here are some of the basic
ideas:

Art is driven by unity and variety. Music, literature, painting, etc
all demonstrate a preoccupation with unity and variety. The
perceptual clockwork that parses data in terms of unity and variety is
surely the same clockwork driving concept formation. If you think, as
I do, that concept formation is THE CENTRAL PROBLEM FOR AI, then art
is the place to look, for art is an endless dance of unity and
variety.

Concept formation is an active area of AI research. There are two
prominent approaches, the classical view and the conceptual clustering
view. The classical view is driven by measurements of similarity.
The conceptual clustering view augments similarity measurements by
attempting to fit data to pre-existing concepts, sometimes contrary to
the groupings that similarity measurements alone would produce.

Art embodies both approaches. Unity based on similarity is obvious.
But the artist's eye also parses data based on pre-existing concepts.
If a pattern of shapes, for example, is roughly "square-ish" we will
parse the image based on the concept of a square.

Interestingly, clustering based on similarity and clustering based on
pre-existing concepts are both quite arbitrary. The groupings
produced depend entirely on what qualities you select as a measurement
of similarity or what concepts you select for your data fitting. The
guiding principle here is utility. The correct clusterings are the
ones that "work". It's not a matter of a concept being true or false;
concepts are simply useful or not useful. The concepts that you wind
up with are determined by what you want to do.

But astonishingly art has another agenda. Art is interested in not
just slicing the conceptual pie into pieces, but in slicing the pie
into "interesting pieces". What could possibly be going on here?

First, we should acknowledge that whatever is going on, is not
pointless or arbitrary. Our large brains and nervous systems are
biologically expensive. They consume approximately 25% of the body's
energy while accounting for only 1% - 2% of the body's weight. James
Anderson has pointed out that if it were possible to reduce the energy
consumed by our nervous systems, "strong selective pressure would have
accomplished it long ago." It's a tough world out there, and
evolution has a relentless way of eliminating the irrelevant. We can
safely assume that aesthetic preference serves a function; we can
safely assume that aesthetic preference serves a function with respect
to concept formation. It is no accident that we are the only species
to develop both high level concept formation and aesthetic
sensitivity.

Now here is the payoff. Concept formation guided by the aesthetic
preference for unity and variety MUST be uniquely effective AS AN
INFORMATION PROCESSING TECHNIQUE. We did not evolve aesthetic
sensitivity so that we can match the sofa with the drapes. Slicing
the conceptual pie into "interesting pieces", is, in all probability,
the most effective way to do it from the point of view of subsequent
information processing. Exactly what "interesting pieces" means and
how it relates to effective information processing are left as an
exercise for the reader.

_________________
8/19/04

A few additional thoughts:

In the previous post I suggested that the mechanisms of art are
closely related to concept formation. Art objects display a high
degree of unity and variety. Unity is just another name for
clustering based on some similarity measurement. Variety just means
you have a collection of these clusters whose similarity measurements
vary greatly. Art IS concept formation.

Actually, that's not quite right. Rather, art is based on the same
perceptual/conceptual mechanisms that drive concept formation. Art
and concept formation are isomorphic.

So when I argue that, "Concept formation guided by the aesthetic
preference for unity and variety MUST be uniquely effective AS AN
INFORMATION PROCESSING TECHNIQUE", I don't mean to imply that concept
formation is literally guided by aesthetic preference. Rather,
concept formation is driven by the same perceptual/conceptual
mechanisms which subsequently inform art.

Okay, if art and concept formation are, in fact, isomorphic, what can
we learn? One of the most striking things to me is that art objects
are SYSTEMS of interrelated elements. Isomorphically, this suggests
that the brain is in the business of storing systems of interrelated
concepts. Individual concepts may not even exist in any meaningful
sense, only systems of concepts. This would certainly make sense from
an information retrieval point of view; you want to store related
elements "close together."

I think, clearly, it is time for AI researchers to dust off the
writings of the gestalt psychologists and take another look. Art
objects are definitely gestalts. Isomorphically the brain probably
structures its contents in terms of gestalts. (the gestalt
psychologists reading this now are gently chuckling at this point) But
it is incumbent upon us, people interested in modeling intelligence,
to determine why from an information processing point of view this is
a truly excellent way of doing things. For it assuredly is an
excellent solution, perhaps the optimal solution. Efficient
information retrieve is undoubtedly just one of many factors here.

I'm looking for an algorithm. It is our species most powerful tool.
We use it to structure our concepts. It created the Calculus, quantum
physics, Chartres Cathedral and the Mass in B minor. It enables us to
cross the street and to cross the solar system. We use it to built
bombs and to build housing.

A fascinating aspect of this algorithm is its ability to construct
entirely new concepts. It structures data into similarity grouping
which can be efficiently stored and manipulated. If these similarity
groupings prove to have utility, if they help us keep the wolf from
our door, they will be given labels and recognized as concepts. This
algorithm creates the new, or rather it creates the concepts with
which we recognize and manipulate the new.

The creation of new concepts has been a tough problem for AI. It is
sometimes called the "problem of terms". Art, in my opinion, will
teach us how to parse data so that new terms can be recognized,
conceptualized and subsequently manipulated in the most efficient
manner.

The algorithm, by the way, has a name, derived from its ability to
create new concepts. It's called the Bo Diddley Algorithm. The
Rock-n-Roll pioneer, Bo Diddley, was fond of saying, "Everything I
know, I taught myself".

:-)

_________________
8/20/04

"Sergio Navega" <snavega@intelliwise.com> wrote in message
news:<41227628_3@news.athenanews.com>...

>> First, we should acknowledge that whatever is going on, is not
>> pointless or arbitrary. Our large brains and nervous systems are
>> biologically expensive. They consume approximately 25% of the
body's
>> energy while account for only 1% - 2% of the body's weight. James
>> Anderson has pointed out that if it were possible to reduce the
energy
>> consumed by our nervous systems, "strong selective pressure would
have
>> accomplished it long ago." It's a tough world out there, and
>> evolution has a relentless way of eliminating the irrelevant. We
can
>> safely assume that aesthetic preference serves a function; we can
>> safely assume that aesthetic preference serves a function with
respect
>> to concept formation. It is no accident that we are the only
species
>> to develop both high level concept formation and aesthetic
>> sensitivity.

>I suggest that the correct interpretation may be the exact opposite
>of the above. It doesn't seem an important evolutionary advantage to
>be able to produce art. But it is important to have a powerful brain,
>capable of elaborating complex models of its world and also able to
>collaborate (and communicate) with peers. Perhaps what caused the
>extinction of Homo Neanderthalensis was a deficiency in the latter.
>Art seems to be a side effect of our huge (and energetically
expensive)
>brain.

Right, art is driven by the same conceptual/perceptual mechanism
which, in my opinion, drives concept formation. The two are
isomorphic as I argue in a subsequent post. And I agree that there is
no evolutionary advantage in being able to produce art. The
evolutionary advantage lies in theconceptual/perceptualmechanism which
is responsible for both art and concept formation.

>> Now here is the payoff. Concept formation guided by the aesthetic
>> preference for unity and variety MUST be uniquely effective AS AN
>> INFORMATION PROCESSING TECHNIQUE. We did not evolve aesthetic
>> sensitivity so that we can match the sofa with the drapes. Slicing
>> the conceptual pie into "interesting pieces", is, in all
probability,
>> the most effective way to do it from the point of view of
subsequent
>> information processing. Exactly what "interesting pieces" means
and
>> how it relates to effective information processing are left as an
>> exercise for the reader.

>The fact is that much of what appears to be the "interesting pieces"
>is heavily influenced by current culture. Art changes a lot
throughout
>time, what remains constant is our ability to manipulate complex
>abstractions and sophisticated analogical mappings.

What I am calling "interesting pieces" has nothing to do with culture.
 The
pieces are interesting because of their form rather than content.
Culture would have a lot to say about content. Our perceptual
mechanisms dictate what is interesting from a purely formal point of
view.

Regards,
Glen

_________________
8/20/04

"Glen M. Sizemore" <gmsizemore2@yahoo.com> wrote in message
news:cg4mtk$1ddl$1@news.wplus.net..

> GS: Translation: the ability to "do art" relies on the same behavioral
> mechanisms involved in more mundane sorts of behavior (concept formation is
> simply an aspect of stimulus control). Art did not arise as a specific
> "module" but, rather, is a result of widespread behavioral mechanisms that
> evolved because animals whose behavior could change as a result of contact
> with the environment were more fit.

Art is a large subject. It has a significant cultural component, and
it has a significant component based on content. I'm not addressing
either of these. I'm talking about art purely in terms of form. Form
has nothing to do with behavior.

Literature and the visual arts complicate the issue by being heavily
involved with content. We can simplify the discussion by focusing on
instrumental music, the Beethoven piano sonatas, for example. The
have no content; they are purely formal.

It is my argument that these pieces are a delightful interplay of
unity and variety, and they are NOTHING but a delightful interplay of
unity and variety. This is not my idea. Any text on the psychology of
art will say the same thing.

I am simply pointing out that unity, the parsing of data into
similarity groupings, is one of the central techniques of concept
formation. Variety arises when you juxtapose groupings with widely
varying measurements of similarity.

My argument is that art objects and concept formation are isomorphic.
They certainly both involve similarity grouping, and they probably
also both involve variety. Variety is an obvious component of art;
let me describe how I believe variety is also an essential component
of concept formation.

The gestalt psychologists have been arguing for some time that we
perceive and understand the world, not in terms of individual
elements, but in terms of systems of related elements. Everything
affects everything else. They are clearly correct.

For the brain to model reality in this manner it must store concepts,
not as individual, isolated elements, but as components in a larger
system. Concepts themselves do not demonstrate variety. It is these
systems of interrelated concepts which contain the variety.

Art is isomorphic with these systems of related concepts. What I'm
looking for here is the way these systems are structured. They are
clearly an excellent solution in terms of information retrieval,
analogy formation, etc. etc. The point of this thread is that by
studying the purely formal aspects of art we may gain insight into
this essential structure.

*****************************************************************
* Glen Foy
* www.clairvaux.org
* "When all you have is a hammer, everything looks likea nail."
* -- Conceptual Clustering
****************************************************************

Quantcast