X-Message-Number: 7552
From:  (Thomas Donaldson)
Subject: Re: CryoNet #7542 - #7545
Date: Thu, 23 Jan 1997 23:36:30 -0800 (PST)

To Mr. Coetzee:

A great deal of work has been done on memory since Pietsch wrote his book.
If you want a good summary of it, I would suggest Steven Rose, THE MAKING OF
MEMORY: FROM MOLECULE TO MIND. I have written about the research going on
at length in my newsletter PERIASTRON.

In one sense Pietsch was as correct as he might have been at the time. Right
now the consensus among neuroscientists is that our brains form one or more
neural nets, with a special structure unlike that of the hardware neural 
nets of computing (memory in a neuron is stored by an increase in strength
of a synapse, OR growth of that synapse, OR formation of a new synapse. All
have been observed to happen in animal brains after learning. The latter two
do not occur in computer neural nets, for obvious reasons. (Someday we might
design nets which do this, but that's a separate issue).

Some other papers relevant to this issue are:
MG Stewart, SN Patel, "Changes in the number and structure of dendritic
spines 25 hrs after previous avoidance training..." BRAIN RESEARCH 463(1987)
  168-173 --- to understand this paper you must understand what a dendritic
  spine is. Particularly when active, synapses are at the top of a spinelike
  protrusion from the dendrite. 
MG Stewart, "Morphological correlates of long term memory in the chick
  forebrain consequent on passive avoidance learning" in LR Squire,         
  R Lindenlaub (eds) THE BIOLOGY OF MEMORY, Schartauer Verlag 1990
M Schechter, D Muller et al NATURE 372(1994) 777-779
WT Greenough, "Structural correlates of information storage in the 
  mammalian brain: A review and hypothesis" in TRENDS IN NEUROSCIENCE
  7(1984) 225-233
K Wu, IB Black, "Regulation of synaptic molecular architecture in a rat
  sympathetic ganglion and hippocampus", J COGNITIVE NEUROSCIENCE 1(1989) 194-
  200.

There are others. I say that he was "as correct as he might have been" 
because when he wrote neural nets were under a cloud. They do share one
trait with holograms: destruction of a proportion of the neurons in such
nets (including the hardware nets used in computing) will leave the memory
they contain more fuzzy, but still there. In that sense they can serve
equally well to explain Pietsch's observations. 

As for a hologramic theory in any strict sense, no, no one has bothered
(so far as I know) to refute it. This is (I think --- it's not that I've
gone out and taken a poll) thought to be so wildly unlikely by neuroscientists
working on memory that it's not worth the ink for refutation. Such a theory
must solve several problems that do not arise for a neural net theory:
first of all, just how would our brains read in and read out this 
hologram (remembering how holograms are created in physics)? The neural
net theory provides a means for read-in and read-out. In the form of 
hardware neural nets, it even gives us hints about how our brain behaves.

I hope that this will be satisfactory. Ideally, you might read Rose's
book, but if you don't want to do that then too bad. (After his book you
can try the papers, and I can give you more if you want).

There's a lot we still don't understand about memory, particularly the
really long term kind. One thing which has not been done is to look at
all the different kinds of memory (which seem, each kind, to center on
particular brain regions) to see how well this theory matches each
kind. (It may need modification --- or perhaps the differences come
only from difference in connections, which can affect the behavior of
hardware neural nets also). The chemistry of individual synapses, 
even in our hippocampus, varies with region in our hippocampus, but
the general pattern remains.

When I said that we are apparently a collection of neural nets rather
than a single one, I was referring to the fact that different activities
have centers in different areas of our brain. Also, neural nets are
very highly parallel computers (made up out of individual pieces which
are comparatively small), while our sense of consciousness and awareness
is sequential. Although this issue is not so well understood, it's 
known from experimental studies that when we attend (say) to something
visual, the relevant neurons in our occiputal cortex (at the back of 
our heads. Don't ask me how our vision is managed at the back of our
heads rather than just behind our eyes!) synchronize their firing.
That is, in a sense, how "read-out" works in our brain.

			Best and long long life,

				Thomas Donaldson


Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=7552