mmm@cup.portal.com (Mark Robert Thorson) (11/12/90)
I'm not in the AI field -- so I could be all wrong about
this -- but it seems like the "fluid and crystallized
intelligence" meme hasn't yet made the jump from the IQ-
testing folks to the AI community. I'll try to correct
that situation in this posting, then go on to make wild
speculations about the underlying mechanisms represented
by these important concepts.
Our story begins with factor analysis, a powerful and
controversial statistical technique. It is a technique
for determining the minimum set of sources of variance
that can account for the variance observed in a body of
data, such as IQ test scores. The most significant
discovery to emerge from factor analysis of IQ test data
is that there are two types of intelligence, for which
Catell [1] coined the widely-accepted terms "fluid" and
"crystallized".
Crystallized intelligence consists of acquired skills
and draws upon learned knowledge. Examples of tests
with a high correlation to crystallized intelligence are
vocabulary tests, math tests, etc. Fluid intelligence
is marked by problem-solving and the perception of
relationships. Tests with a high correlation to fluid
intelligence include predicting the next element of a
series of numbers or figures, spatial visualization, and
word analogies.
When the two factors are separated, it can be seen that
scores on tests for fluid intelligence track together as
a group, indicating a single quantity is being
represented. Likewise, tests weighted toward
crystallized intelligence track together as a separate
not-quite-independent group. (Critics point out this is
the inevitable result of using factor analysis to define
your groups, but the existence of these factors has been
confirmed by experimental data.)
A person high in fluid intelligence is also likely to be
high in crystallized intelligence, but this is not
always the case. It appears that a person's fluid
intelligence can be spent on things -- such as athletics
or art -- which are forms of crystallized intelligence
not measured by IQ tests.
Tests of crystallized intelligence show a gradual
increase from birth to old age, with a decline in the
rate of increase during the teenage years. Tests of
fluid intelligence show a peaking in the late teens or
early twenties, with a gradual decline that accelerates
around age 55 or 60.
Although IQ tests can't show whether fluid intelligence
becomes crystalline intelligence, that is my opinion as
illustrated below.
* - * * * - * - * * *
| | | | | | |
* - * - * - * - * * - * - * - * - *
| | | | | |
* * - * * * - * - * - *
| | | | |
* - * - * - * - * * - * - * - * - *
| | | | | |
* * - * * - * * * - *
| | *
* * *
*
*
* * *
* * *
The upper part of the diagram represents a crystallized
network of agents. The bottom part represents fluid
intelligence, which I speculate are a pool of
uncommitted agents. These pools exist in every mental
organ of consciousness.
Crystallization occurs when a fluid agent is recruited
into the network. This occurs whenever you have a new
thought. It can be a temporary recruitment such as
deciding to have a drink of water, or it can be a
permanent recruitment like deciding not to smoke
tobacco. Temporary agents return to the pool soon after
they become obsolete, because lack of use softens their
connections. The permanent agents tend not to return to
the pool either because frequent use (i.e. interrogation
from the network caused by re-thinking the thought)
hardens their connections, or because the development of
additional connections binds them more tightly to the
network, or because they become buried in the network by
the recruitment of subsequent agents.
Occasionally defects, such as folds or bifurcations,
occur in the pattern of growth of the network, as shown
below.
************* *************
*********** *************
*********** *************
*********** ***********
********* ***********
********* ***** *****
******* **** ****
***** **** ****
* ** **
Normal Bifurcated
This can be a benign occurrence, as in the case of the
development of the language organ in bilingual people.
Some agents are common to both languages, but there are
two distinct subpopulations or lobes of language-
specific agents. The agents in these lobes mostly have
connections among themselves and to the common base,
rather than to the other lobe.
This can also be a dysfunctional condition, resulting
from a trauma at a critical period during development
and resulting in a mental disease such as -- in the
extreme case -- multiple personality disorder. Each
lobe of the bifurcation becomes a separate and competing
suborgan, reinforcing its own connections and
exacerbating the problem.
This perhaps explains why electroconvulsive therapy
(ECT) needs to be given in several rounds over a period
of weeks to be effective. Each round peels back another
layer of crystallized intelligence and makes the layer
below it more vunerable to the next round. Eventually
the root of the bifurcation is reached, and the network
can be allowed to re-form or re-grow or re-organize or
anneal, as shown below.
************* ************* *************
************* *********** ***********
************* *********** ***********
*********** ********* ***********
*********** --> ***** --> *********
***** ***** *********
**** **** *******
**** **** *****
** ** *
Before ECT After ECT Months later
Why is the distinction between fluid and crystallized
intelligence important? I believe you cannot have
consciousness without having both. If you only have
crystallized intelligence, you merely have a clockwork
mechanism. Such a mechanism could re-think old
thoughts, but it could not have any new thoughts. It is
the having of new thoughts -- through the conversion of
individual particles of fluid intelligence into a
network of crystallized intelligence -- which is
consciousness.
------------------
[1] Catell, Raymond B., "Theory of fluid and
crystalized intelligence: A critical experiment",
Journal of Educational Psychology, 1963, 54, 1-22.jpk@ingres.com (Jon Krueger) (11/13/90)
From article <35870@cup.portal.com>, by mmm@cup.portal.com (Mark Robert Thorson): > Each round [of convulsions] peels back another > layer of crystallized intelligence ... If you assert this, you can hardly deny that injecting dye into CNS tissue causes colorful thoughts. -- Jon -- Jon Krueger, jpk@ingres.com
mmm@cup.portal.com (Mark Robert Thorson) (11/19/90)
I think neural plasticity must be limited to that degree which keeps us from getting too screwed up. For example, it would be disadvantageous to learn some quirky, non-optimal way of walking -- or not to learn to walk at all. The fact that a few people become cross- eyed or autistic or dyslexic points out the hazards of too much neural plasticity. So we must be born with crystallized low-level sensory and motor agencies. Indeed, a newborn baby will perform walking-like motions if held upright over a surface it can reach with its feet. Move it forward along the surface, and it will raise and lower its legs as though it is walking. This ability disappears rapidly, and re- emerges months later. Natural selection must have programmed us with _how_ to walk. It is _when_ to walk which is controlled by the crystallized agencies which develop after birth. Most of the nervous system is concerned with crystallized intelligence. Some inputs are processed to outputs solely through crystallized agencies, such as the pathways which control breathing and balance. Others have crystallized sensory front-ends, such as the visual and speech recognition systems. They also have crystallized motor back-ends, such as the speech and walking motor pattern generators. But in the middle sits consciousness -- a mix of fluid and crystallized intelligence which has thoughts and makes decisions -- as shown below. ---->************-* * ************----> I ---->************ *-*-* *-*************----> O n ---->************ | | | ************----> u p ---->************-* *-*-* ************----> t u ---->************-* * ************----> p t ---->************ * *-************----> u s ---->************************************----> t ---->************************************----> s ---->************************************----> The pool is like the gap of a spark plug. When the voltage across the gap gets high enough, some of the air in the gap ionizes. Soon, a thread of ions bridges the gap, and a small electrical current flows across. The initial trickle of current initiates a self-reinforcing feedback loop, in which more current ionizes more air, lowering the electrical resistance of the gap and increasing the current. Likewise, when a thread of crystallized agents forms between a recognized sensory event and an appropriate response, that thread gets reinforced. Reinforcement yields more use and more reinforcement, which hardens the thread making it a permanent addition to the network of crystallized intelligence. In my previous posting, I speculated that there are pools of fluid intelligence in each mental organ of the conscious brain. Actually, I think that fluid intelligence resides not _within_ an organ, but _between_ organs. This seems obvious when you consider that the connections within an organ need very little neural plasticity. An organ is born with its function mostly intact. It is the control paths between organs which require plasticity. For example, the visual system has a built-in array of face-recognizers, but it is the connections between the face-recognizers and the action- generators which make the decision to run from the face of an enemy. In this model, the low-level sensory and motor organs are capped by an interconnecting network of pipes. At birth, these pipes are clear fluid intelligence. But with use, a web of crystallized intelligence builds up in each pipe. Consciousness is the interaction between the web and the fluid, as new threads form and old ones either become permanently hardened or dissolve back into the fluid. My model of the organs of consciousness is too complex to express in ASCII graphics, but if you send me an SASE, I'll send you a free copy of it. This is the same map of consciousness I distributed last year. I would like to hear any comments on improving this map. Mark Thorson 12991 B Pierce Rd. Saratoga, CA 95070 USA
jpk@ingres.com (Jon Krueger) (11/20/90)
From article <36059@cup.portal.com>, by mmm@cup.portal.com (Mark Robert Thorson): > Most of the nervous system is concerned with > crystallized intelligence. Sure. And most of your TV's design is concerned with optimal reception of the Brady Bunch. That's why there's a special "Marcia detector" circuit to help her find guys. Also you'll find the "Alice and Sam feedback oscillator" that detects when Alice is depressed and prompts Sam to appear. You believe that, don't you? No? Then why do you persist in claiming that neurons have ideas? -- Jon -- Jon Krueger, jpk@ingres.com
mmm@cup.portal.com (Mark Robert Thorson) (11/21/90)
> appear. You believe that, don't you? No? Then why > do you persist in claiming that neurons have ideas? I thought I was careful to use the term "agent" when talking about individual particles of intelligence. This could be a higher-level entity than a neuron. As an analogy, the gates and flip-flops of a computer are not made from individual transistors -- they are clusters of transistors, each serving a different function within the unit. Likewise, I think that the different types of cells in the cerebral cortex may serve specialized functions in the complexes of cells which are the atoms of intelligence. If 1 agent == 1 neuron, it would be more likely that the cortex would be composed of a single cell type, which it is not. Another piece of evidence supporting the 1 agent = many neurons concept is the structure of the cerebellum. This organ is a repeat structure of the same complex of neurons across its entire surface. I forget the number of neurons in the complex, but I think it's about 15. There's a hierarchy of structure to the human brain, which is mostly unrecognized due to the profound influence of the Golgi stain on neurophysiology. The Golgi stain is a technique for visualizing individual nerve cells in a tissue. For unknown reasons, the Golgi stain only affects about one cell in a thousand -- seemingly a cell picked at random. This makes the cells more visible, because the darkly-stained cell stands out against the mostly-colorless unstained cells. This allowed the early neuroanatomist Santiago Ramon y Cajal to produce his stunning series of drawings describing most of the major cell types in the nervous system. These drawings not only appear in many texts, but have influenced succeeding generations of anatomical illustrators and photographers. It has programmed an unconscious bias. A drawing or a photo that doesn't look like a Cajal drawing looks subtly wrong. Unfortunately the Golgi stain has exerted a bias on our view of the brain. We know a lot about the individual cell structures, but the intermediate-level structures are completely invisible to Golgi staining. It's as though the only way to look at a tree is to cut down the neighboring forest for the radius of a mile.
jpk@ingres.com (Jon Krueger) (11/22/90)
From article <36099@cup.portal.com>, by mmm@cup.portal.com (Mark Robert Thorson): >> why do you persist in claiming that neurons have ideas? > ...I think that the > different types of cells in the cerebral cortex may serve specialized > functions in the complexes of cells which are the atoms of intelligence. Do you believe that if we put enough books in the library it will start having ideas too? Having lots of things of one sort doesn't automatically turn it into a thing of another sort. -- Jon -- Jon Krueger, jpk@ingres.com
powers@uklirb.informatik.uni-kl.de (David Powers ) (11/23/90)
jpk@ingres.com (Jon Krueger) writes: >Do you believe that if we put enough books in the library it >will start having ideas too? Yes! If you represent the books and the library the right way. >Having lots of things of one sort doesn't automatically turn >it into a thing of another sort. See the discussions on emergence and self-organization. If the units have the right potential for relationships, significant properties can emerge through self-organization (or other effects). It's started to snow here today - so I'll take that as an example. Have you looked at a snow flake recently - self-organized from a collection of water molecules (cf "books"), in the right environment (cf "library"). David ------------------------------------------------------------------------ David Powers +49-631/205-3449 (Uni); +49-631/205-3200 (Fax) FB Informatik powers@informatik.uni-kl.de; +49-631/13786 (Prv) Univ Kaiserslautern * COMPULOG - Language and Logic 6750 KAISERSLAUTERN * MARPIA - Parallel Logic Programming WEST GERMANY * STANLIE - Natural Language Learning Riddle: What is the difference between the university and me. Disclaimer: My opinion.
mmm@cup.portal.com (Mark Robert Thorson) (11/27/90)
Language is first and foremost the reproduction
mechanism for memes. A meme (as defined by Dawkins in
_The_Selfish_Gene_) is an individual particle of
culture, just as a gene is an individual particle of
genetic inheritance. A meme can be an individual
thought, such as "soup is good food" or a complex of
thoughts, such as a religion.
Language is how memes jump from person to person. On
the transmit end, the language organ photographs a
portion of the network of crystallized intelligence and
encodes it for shipment. On the receive end, it decodes
the package and temporarily installs it into the
corresponding place in the receiver's network. The code
is free of the context of the transmitter and receiver.
For example, if I say "This tastes like fish." that
would seem like a good thing to a receiver who likes
fish, or a bad thing to one who doesn't. I.e. the
message doesn't carry the context with it unless it is
specifically encoded, as in "This tastes like -- yuck --
fish.". (Voice inflection can transmit the same
information, but that is just another form of specific
encoding of the context.)
I can tell you anything in my conscious mind, from why I
don't eat pickles to why I don't go to church.
Likewise, I can input almost any idea from you. I might
not agree with the truth or falseness of what you say,
but I can try it on for size. I can map it into my
network and see if it fits.
Somehow, the language organ is like some sort of robot
arm, with random-access pick-and-place reach into
arbitrary places in the network, as illustrated below.
* - * * - * - * - * * - * - * * - * - *
| | | | | | | |
* - * - * - * * * - * - * - * - * * - *
| | | | | |
* - * - * - * * - * - * - * - *
| | |
* * - * *
(*)
-------------- /\ V
| Language | / \ |
| Encoder/ |_/ \|
| Decoder |
--------------
How could an organ have such all-invasive access? It
could selectively activate individual particles of
crystallized intelligence using an address bus. For
example, when the address bus for the food department of
my brain carries the code for "pickles", an address
decoder activates my pickle-agents including one
connected to my "too much salt" agent. Seventy binary
signals can address more than a billion billion
particles, so obviously such an address bus needn't be
unreasonably large. (It would probably be much larger
than seventy binary signals, however, in order that a
random address picked out by an agent would be likely to
be globally unique, much like the system used to assign
credit card numbers.)
My guess is that the language organ has two parts: a
centralized encode/decode part (Broca's and Wernicke's
areas, etc.) hooked up to the hearing and vocal organs,
and a distributed part -- the "robot arm" -- consisting
of one or more sparsely encoded buses capable of
interrogating all of the conscious agents and agencies
and even placing new agents and constructing new
agencies, although the newly-arrived memes seem to have
weak connections, and require reinforcement from the
existing network to become permanent. (I.e. you are
much more likely to believe your own conclusions than
those spoken to you or read in a book, until you've had
time to consider them.)
In an earlier posting, I claimed that thoughts are the
experience of agents crossing the fluid vs. crystallized
interface. Now, I'd go further and claim that dreams
are the experience of agents and agencies spontaneously
forming and re-dissolving while the robot arm is idle.
We don't perceive dreams while we're awake because
agents and agencies constructed by the arm are formed at
a higher voltage or pressure or something. While the
arm is active, we don't see the spontaneous activity,
just as we don't see the stars when the sun is out. The
higher intensity of the connections created by the arm
is lacking in the agents and agencies formed during
dreams, which is why dreams are forgotten so quickly.
New thoughts begin in this haze of spontaneous activity
(which is always present, even though we only perceive
it at night). A new thought occurs when two
crystallized agents need a connection, and a fluid agent
jumps into the gap. If the connection is really needed,
it gets reinforced and becomes permanent.ffujita@s.psych.uiuc.edu (Frank Fujita) (11/27/90)
In article <36244@cup.portal.com> mmm@cup.portal.com (Mark Robert Thorson) writes: >Language is first and foremost the reproduction >mechanism for memes. Language is many things, among which it is also... Now that we have that right, back to your regularly scheduled newsgroup.
sena@infinet.UUCP (Fred Sena) (11/28/90)
In article <36059@cup.portal.com> mmm@cup.portal.com (Mark Robert Thorson) writes: >My model of the organs of consciousness is too complex >to express in ASCII graphics, but if you send me an >SASE, I'll send you a free copy of it. This is the same >map of consciousness I distributed last year. I would >like to hear any comments on improving this map. > Could anyone out there recommend some good books that explain models of consciousness? I'd like to find some sources that don't require a lot of AI background to understand. --fred -- -------------------------------------------------- Frederick J. Sena sena@infinet.UUCP Memotec Datacom, Inc. N. Andover, MA