From newshub.ccs.yorku.ca!torn!cs.utexas.edu!uwm.edu!linac!mp.cs.niu.edu!rickert Wed Aug 12 16:52:13 EDT 1992
Article 6552 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!torn!cs.utexas.edu!uwm.edu!linac!mp.cs.niu.edu!rickert
>From: rickert@mp.cs.niu.edu (Neil Rickert)
Subject: Re: Memory and store/retrieve.
Message-ID: <1992Aug3.200351.3632@mp.cs.niu.edu>
Organization: Northern Illinois University
References: <1992Jul30.152320.2247@puma.ATL.GE.COM> <1992Jul31.160209.26718@mp.cs.niu.edu> <1992Aug3.151610.21034@puma.ATL.GE.COM>
Date: Mon, 3 Aug 1992 20:03:51 GMT
Lines: 89

In article <1992Aug3.151610.21034@puma.ATL.GE.COM> ljones@andrew.ATL.GE.COM (LeRoy E Jones) writes:
>In article <1992Jul31.160209.26718@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil Rickert) writes:
>>
>>  But how exactly do you implement this decision to attempt to remember?
>>The usual method is for you to try to retain the information in your
>>conscious thoughts for as long as possible.  This is highly consistent
>>with the idea that learning is an accretion process, and that by keeping
>								  ^^^^^^^
>>the information in your thoughts you are lengthening the time over which
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>the accretion occurs.
>
>How does one keep information in his/her thoughts? Without a store function,
>at least for short term memory, the only way to do this is to continually
>observe the input, but the input couldn't change, and would need to be 
>available "long enough."

  As I've mentioned, I see memory being primarily organized towards
recognition, probably based on partial clues.  I see thought and
consciousness as a simulation process.  The simulation is maintained
such that it can be recognized by the memory recognition process.  There
will, of course, be some tendency for the simulation to drift from the
original sensory input, thereby reducing the accuracy of memories.

  Incidently, I see consciousness as having evolved primarily for the
benefit of strengthening important memories by such a simulation.

>Somewhere along the line, I fit the evidence you submitted for accretion/
>inference, into another model utilizing storage/retrieval and explained
>how symbolic links between information units can acount for the observed
>phenomena with memory. If you remember when I put this forth, what was your
>objection to it?

  I guess I'm troubled by its apparent need for an overall top down
design.  Somebody has to say "lets start evolving features so as to
prepare a future linked list arrangement."  But I see evolution as
primarily bottom up.  Features are enhanced as they provide benefits.
This being said, it is possible that something could evolve in a bottom-up
design, yet the resulting effect be interpretable as an apparent linked
list arrangement.

>Also, you mentioned specialized memory for things like language. What is
>different about our memory of words and such?

  I don't think I referred to specialized memory.  Rather I suggested
that the special nature of language affects the way memory is used.
The important point is the binary nature of language.  This makes it
resistant to signal degradation.  A language signal can degrade
significantly, yet still be accurately recognized and recreated.  As
with digital telephone communications, compact disks, and other digital
techniques, this allows signals to be resistant to degradation.  An
important consequence is that memory refresh becomes possible.  The use
of simulation through thought to sustain memory and reinforce learning,
as described above, will resist degradation for linguistic thought,
whereas for non-digital information the simulation will tend to slowly
drift from the original.

>Do you think we infer even the most basic facts, like letters in the alphabet,
>digits in the number system, or that 4 is the answer to the question "What
>is 2 plus 2 ?"

  I see "2 plus 2 equals 4" as a basic pattern we learned to recognize
in elementary school.  With sufficient clues our thought processes can
regenerate that pattern, and match it against memory to ensure it matches
what we can recognize.  Thus "2 plus 2" should be enough of a clue to
recognize, and prompt the regeneration, through thought, of the complete
pattern.

>               Assuming your inference model, it seems that with continual
>use, the inferences would change (I guess increased accretion), and the 
>inferencing is eventually and effectively a retrieval of a stored piece of
>information. 

  Yes.  That would essentially be the case.  As we first learn to
recognize, contextual clues are an important part of the basis for
recognition.  But as we become more familiar with a "fact" we learn to
recognize it in a broader range of contexts.  The broader the context
in which we can recognize, the broader the context in which we can
recreate, and effectively retrieve.

>Why shouldn't we model systems after methodologies which utilize store and
>retrieve paradigms? Aren't they more precise than inference based solutions?

 For our own computing technologies, you are correct.  Store/retrieve
is more precise, and thus is usually better.  But when it comes to
understanding the mind, we must deal with the problem that much information
is quite imprecise.  The weakness of store/retrieve is that it is too
precise, and does not deal well with imprecise knowledge.



