Mark Bieda mbieda at ELROND.STANFORD.EDU
Thu Mar 12 19:25:20 EST 1992

I should apologize for the partial transmission - it was entirely my
mistake... It's also good to see some life in this net...

Replies (certainly not answers) to Tom's questions:

1) Topic 1: The distribution or localization of knowledge
	This is tough one because it matters how you define knowledge.
I think that the current consensus is that there should be a mixed
answer to this - it is quite clear that certain abilities, such as the
ability to recognize an object presented visually, are localized to
fairly specific regions of the brain. However, within that general
area, the mode of knowledge representation is anything but clear.
Perhaps the most important point is to realize that the way we use the
word 'information' or 'knowledge' is very different from the way the
brain uses it. The brain handles very different sorts of knowledge in
very different ways. Motor programs are handled differently from
visual knowledge, etc. See the end of this note for a rationalization
of this circumstance.

2) Sections of the brain and activation for various tasks... 
   This goes back in most ways to the answer to question one. For input
   (vision/audtion/olfaction/ect.) and output (moving a leg, moving the
   eye toward a specific object, ect.) we do know of specific areas of
   the brain that are specialized, even dedicated to these functions.
   However, for the larger questions of putting it together, we are
   almost completely in the dark.

3) Artificial Neural Networks and the Brain.
   This is a tough one. As a member of the Rumelhart lab, a neural
   network lab, and the Tsien lab, a physiology lab, I tend to straddle
   the fence on these questions. Again, the philosophical questions are
   difficult and I think necessary to approaching this question. I
   believe that neural networks to some extent reflect the computational
   style of the human brain. Although this may seem radical to some, I
   think that it is almost a truism... the brain clearly uses massive
   parallel processing, at least for its basic functions. Now, let's
   accept that the brain is, to a large extent, parallel and
   distributed. This does not imply that the computational strategies
   used by the brain are very similar to those used by neural networks.
   But the jury is still out. You will find a large volume of work on
   this subject. Francis Crick wrote a harsh opinion piece on neural
   network work for Nature or Science a while back that is still valid
   today. And various others have commented on these issues. For your
   purposes, it probably doesn't matter how closely the nets mimic
   actual brain computation - although knowing more about how the brain
   computes may help you increase the power of your nets. That is
   certainly an active area of investigation here at Stanford. As for
   your question about brain simulation by computer... most people I
   know think that it will take thousands of years to understand the
   brain. There is no adequate simulation of any single neuron on
   computer anywhere!

4) Feedback and the brain.
   There is feedback all over the brain. Why? Well, we can make some
   general statements in some cases, and actually say something specific
   in a few cases, but mostly we are in the dark. I think that this is
   arguably the least well understood major feature of brain processing.
   Oh, and feedback is absolutely fundamental to sensory processing,
   beginning at very low levels, in most cases that of the actual
   sensory receptor cells.

Vague questions:

5) Macroscopic and Microscopic View.
   An area of more heat than light, and simply not much heat or light
   recently. I think that the prevalent view is that low level
   properties of neurons will play important roles in structuring
   computation. That wonderfully vague statement leaves open a huge
   realm of possibilities. We simply have little ability to test these
   possibilites right now. I tend to see these larger questions break
   down into smaller questions that are more tractable. So let's see -
   the cerebral cortex has a basic 6 layer structure, and neurons of a
   certain shape, called pyramidal cells, tend to provide the output
   from one area of cortex to the next. Now, Brodmann in the early part
   of this century found that cortex can be divided into different areas
   by the "look" of cross sections. In some areas, there is a big layer
   four, for example. Well, let's look at everybody's favorite Brodmann
   area - area 17, also known more descriptively as visual area one (V1)
   because it is the first neocortical representation of vision.
   Question: How similar are pyramidal cells of layer four in area V1?
   Well, cells are kind of like individuals, so the real question we
   want to ask is how important are the differences to the processing of

6) reasoning & perception: What are they? 
	Philosophical opinion on these issues has certainly been
informed by the various brain sciences. However, it does remain
philosophical opinion. So talk to a philosopher. Rorty's book
Philosophy and the Mirror of Nature is a good place to start.

General Brain:

	One way to rationalize the chaos of brain organization is to
go back to one of the fundamental guiding principles of biology:
evolution. So here's an analogy that all of us programmers can relate
to. Imagine the brain as starting out, way back, as a simple program
designed to do one simple task. Well, like most effective programs,
after a while there is more demanded of it. But imagine that the
additions are just kind of added on haphazardly - every time a new
task is desired, the code is stuck on in the most convenient (but
least elegant) way possible. We've all dealt with these sort of
programs - five or six people have modified it to do all sorts of
little things, and we end up with spaghetti code. Well, I would argue
that this is how evolution works, in a sense. In an evolutionary
process, you never have time to just start over and rework the whole
system. And evolution is always in a rush, so the additions are
usually made in the most bogus, kludgy way. I think that this
rationalizes why the brain handles different information in so many
different ways, and why there are sometimes torturous paths from point
A to point B - no one in their right mind (bad pun) would ever design
something like this structure! 

	I'm no evolutionary neuroscientist, but for me the structure
of the brainstem, and indeed the course of information in various
sensory pathways (especially audition and vision) seems to follow this
pattern. However, the cerebral cortex is another matter - it does seem
to have similar pathways from place to place.... Could it be that the
cerebral cortex is a radical innovation in brain design that is
superimposed on top of the early developments, in effect a brilliant
new innovation. Inquiring minds want to know.

Well, forgive me if this response has been long winded - your
questions are some of the deepest and most thought provoking in

Recommended readings:

McClelland, Rumelhart and the PDP group. Parallel Distributed
Processing. Vol 1 and 2. The Bibles of the neural network field.
Chapter 20 and 21 cover some basic biology of the brain from a
perspective that you will probably appreciate. However, somewhat dated
now, since the books came out in 1986.

Churchland. Neurophilosophy. A good book, but now seriously dated.
Gives some flavor of the nervous system, but, wow, our whole approach
to the nervous system looks really different now. Also some factual

Kuffler. From Neuron to Brain. A very good introductory textbook.
Quite strong on various issues.

Kandel, Schwartz, and Jessel. Principles of Neural Science. Now
essentially the dominant textbook of neurobiology. Recent edition is
very good, but rough going for those without a biological background.
Also, a bit of a clinical emphasis.

Good luck,

Mark Bieda
Stanford Neurosciences Program

More information about the Neur-sci mailing list

Send comments to us at biosci-help [At] net.bio.net