Hello, everyone, from Bill Powers –
This is getting ridiculous – look at that list in the To field of all
the people who might have something useful to say about some questions I
want to ask. This keeps happening – can’t we all just get along in one
discussion list oriented toward modeling and neuroscience and therapy and
everything else with a technical slant? We did it before – there are
about 130 people in the CSG discussion list, first in the group above.
Does it just get too messy with all the different interests and levels of
expertise, or can we somehow manage to keep it organized by threads?
Martin Taylor had something going – maybe he will have a suggestion. Or
everyone could just subscribe to CSGNET.
Anyway, here is what I want some help with, even though I will have no
idea what mailbox to put the answers in.
Hubel and Weisel found that there were neurons that respond selectively
to visual lines oriented in different directions. Lines in one direction
maximally excite one group of neurons, while physically distinct neurons
light up when lines are in a different orientation. This is very
different from the kind of neural model implied by our PCT diagrams (and
most other control-system diagrams I have seen). Instead of a function
box which indicates direction by the magnitude (frequency) of the signal
it emits, we have different signals each representing one direction
only.
That’s fine if all you have in the visual field are straight lines and
are only interested in controlling their average orientation. What if the
image is a circle? And how the heck do you then represent where
the circle is in the visual field?
A circle has segments oriented in every direction, so in principle
all of those H&W cells should fire when you look at a circle
no matter where in the field it is. A square, on the other hand, would
excite some of the cells less and some more. Do longer line segments give
bigger signals? And how would you discriminate a square and a circle
present at the same time but in different positions in the field, or
concentric around the same point? Or two squares oriented at different
angles? Is there a whole collection of the H&W cells for each point
in the visual field? I should think we would run out of neurons pretty
fast at that rate.
Another question: what does the magnitude of the response from one of
these cells represent, and more to the point, to what system would those
signals be an input? And what would the destination system do with those
signals? Is there some principle of mapping or conformal transformation
or something else esoteric that we need to know about to understand how
this thing works?
It’s very frustrating to keep drawing diagrams and making models with
them that work just fine in predicting tracking and other control
behavior, while knowing, or at least believing temporarily, that the real
system doesn’t do it that way. And I can’t help wondering if there isn’t
something wrong with the H&W-type data or interpretations – are we
just looking at the neural nets in the wrong way? Like from the side
instead of the top, or in the middle of some large computing network
rather than looking at its inputs and outputs? Or is it our silly
preference for a space with only three dimensions that is confusing us?
Do we really, really need to understand what glial cells really
do?
I suppose we can go on modeling as we have been doing, as long as it
continues to work. But it’s going to be difficult to find neural
correlates of control processes if we can’t find one-to-one
correspondences between our diagrams and the actual wiring. We can do
that in the spinal cord – is that as far as this kind of model will let
us go? But we can model visual-motor systems pretty well, and that surely
involves more than the spinal cord.
I think this is a rather big problem and unless we can find some sort of
answer there will remain a lengthy hyphen in the middle of
“NEURO-----SCIENCE.”
Best,
Bill P.