NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/bit/listserv/csgl/1579

Text File | 1992-11-20 | 3.7 KB | 72 lines

Comments: Gated by [e-mail]
Path: sparky!uunet!paladin.american.edu!auvm!BEN.DCIEM.DND.CA!MMT
Message-ID: <[e-mail]>
Newsgroups: bit.listserv.csg-l
Date:         Fri, 20 Nov 1992 14:11:02 EST
Sender:       "Control Systems Group Network (CSGnet)" <[e-mail]>
From:         [e-mail]
Subject: Re: Grandmother cells
Lines: 61

[Martin Taylor 921120 13:40]
(Rick Marken 921120.0800)

Apology accepted, Rick, and no (permanent) offense taken.

>Anyway, the presumed solution to the Grandmother cell storage
>problem -- distributed representation -- is only a solution
>in the minds of the people watching these distributed systems;
>at some point, some resource (a cell or cell assembly) is going
>to have to represent the state of the distributed representation
>to someone other than the person watching the model -- ie. there
>will have to be a perceptual signal that represents the state
>of the distributed representation -- and that requires storage.

This is at the heart of the disagreement between the AI people and the
connectionists, and it is interesting to see it rearing its ugly head
within PCT.  The question in my mind is: why should it be true that
>at some point, some resource (a cell or cell assembly) is going
>to have to represent the state of the distributed representation
>to someone other than the person watching the model -- ie. there
>will have to be a perceptual signal that represents the state
>of the distributed representation...

Why does it need to be represented, within the PCT view of the world?  It
seems to me that all that need happen is that the various perceptual (scalar)
signals within the complex be controlled.  The distributed representation is
a representation of "something" only if there is someone outside that needs
to represent it and (presumably) label it, perhaps not verbally.  That
includes the person whose brain includes the complex, seeing themself
as if from an outside observer.  For acting in such a way as to control the
complex, a single higher-level ECS will not do the job, because the complex
has more than one degree of freedom.  You need many higher-level ECSs, and
then they themselves form a complex.

Sure, you can have "grandmother" ECSs, and as Bill pointed out, specific
face detectors have been found in sheep.  I never argued that you couldn't,
just that for many purposes you needn't.  That's what I meant when I originally
said Bill's point was unnecessary.  (Last night, it occurred to me that
perhaps you thought I meant it was unnecessary for Bill to have made the
point. That would have been discourteous and would have justified the tone
of your original response; I meant it was technically unnecessary.)

------------
I was thinking a bit about conscious perception.  Whatever it may be and
wherever it arises within the hierarchy, it must relate in some way to
perceptual signals.  But it is subjectively very highly multidimensional,
and I don't see how this kind of subjective multidimensionality can be
reconstituted from one scalar signal, no matter how complicated the
function from which that one is derived.  That leads me to think that
at least the conscious aspect of perception must be derived from a complex
of scalar perceptual signals.  If so, then why should the set of perceptual
signals relating to one perceptual "object" such as "grandmother" be required
to be combined into one scalar signal, at any point at all in the process?
Again, it seems necessary to combine them only if you are going to
"represent" the object as something (a symbol, a word, a non-verbal label).

Scalar perceptual signals may well be a fundamental core for perceptual
control, but they don't happen one at a time.  I guess that's the bottom
line for this argument.

Martin