Re: Cool visual illusion

feedbackdroids wrote: 
Wolf Kirchmeir wrote:



In any case, if certain behaviorist strategies work better/worse than
others, then presumably they do so because they implicitly assume
different "mediations" (as Glen calls them), ie different internal
mechanisms. Not that they can provide very fine-grained detail, you're
quite right about that. But they can indicate where to look for that
detail. AFAIK, all studies of the fine structure of the brain begine
with behaviorist observations that suggest a "fruitful area of inquiry",
and all entail behaviorist studies in the experimental design.




Well yes, no one is debating that. However, as pointed out, this only
gets us as far as items #1 or #2 of the list I presented earlier. The
journey has barely begun. What and how are very different problems.

Since Curt never responded to my post from a few days ago, I'll patch
it in, since it's relevant here .... cf, Curt's comment "... Except
that knowledge of the illusion IS knowledge about the internal
mechanisms...".

======================
Curt Welch wrote:

"feedbackdroids" <feedbackdro...@xxxxxxxxx> wrote: 


Visual illusions are, as they say, systems-level problems, and need to
be addressed and analyzed as such. You cannot do it by pure
reductionism, and you cannot do it without some actual knowledge of
"internal" mechanisms.




Except that knowledge of the illusion IS knowledge about the internal
mechanisms.  So once you have that knowledge, you can very much "do it" by
pure reductionism from that knowledge.  The rest of the details about how
the human body creates that illusion is just implementation details.




How do you analyze it? You have a retina with 4 or so types of
photoreceptors, several classes of bipolar and horizontal cells at the
outer plexiform layer, several classes of ganglion and amacrine cells
at the interplexiform layer, and that's just the retina.

Now, add in interactions at the LGN, plus 400,000,000 cells in V1, and
then another 30+ cortical visual areas, with roughly as many cells
each. Then, whatever happens at higher levels in the brain. Then, the
billions of feedback signals between areas, and also from higher to
lower. Then, the various feedback loops between all of these areas and
the parts of the brain below the cortical level. What effect does the
midbrain superior colliculus [ancient visual-movement center] have on
this? Do signals rebound off the cerebellar cells? On and on.


How many total cells are involved in the illusion? Since it covers a
wide area of retina, I'd say at least 5 million retinal cells, and 5
billion in brain, just a guess. How many synapses? Say a trillion.


If starting from ..... "Except that knowledge of the illusion IS
knowledge about the internal mechanisms ...", and given the complexity
of the internal system described, tell me what you can say about those
internal mechanisms with any certainty, without going inside and doing
the measurements. Which of the 40-50 total areas mentioned are NOT
involved in the illusion? What can you really be sure of, when sitting
from without? Not much. 
===============

But sitting from without is all we have. Looking at brain function and physiology merely shifts the locus of the behaviour being studied. Or would you insist that what the brain does is not behaviour? Besides, without that behaviour that is supposedly so uninformative, measurements within the brain are somehat uninformative, too, to put it mildly.

For example, to find out which of the areas mentioned are/not involved in the illusion, you could anesthetise a few clusters of neurons at a time. You might then find that the illusion-generating data takes some pathways and not others - but only if you have the subhect's response to the illusion available while you're doing those measurements. Etc.

I think the guess of "several trillion" cells involved in the illusion is unlikely, or else a red herring. Sure, trillions of cell are involved, but finally most these cells' responses are filtered out. There is increasing evidence that many of the "true illusions" that the VC generates occur in very small clusters of cells, so I would expect the same of the "false illusions." Consider face recognition, for example: recent work suggests that we train appproximately one neuron per face. The effects of strokes also indicate very strongly that many of the higher functions of the brain (the ones naively identified as functions of the mind by many/most people) depend on clusters of a very few cells.

IMO a visual illusion may arise because two or more such clusters are competing with each other, ie, responding to the same inputs differently, and with equal strength. Metaphorically, they report different interpretations of the data simultaneously and with equal probability. Normally, one of the competing clusters responds will report a higher probability than the other(s). But we should not take metaphors too seriously: they are at best indicators of "fruitful areas of inquiry." In this case, one would look for two or more neuron-clusters repsonding equally strongly when an illusion is presented to the subject. The quoted account of the path of nerve impulses from retina to VC do not indicate whether such comparative studies have been done, and I don't know whether currnet technology makes such studies possible.

What visual illusions IMO do confirm is that learning to see X requires that the neurons in the VC a) must form networks; and b) that these networks must be trained to respond to inputs from other areas of the brain (which in turn mediate the sensory inputs which can be manipulated via environmental cues.) Since mamals generally are born with underdeveloped VCs, the infant mammal must learn to see. This learning shapes the brain. Some experiments in the 60s IIRC showed that a kitten could learn to see only vertical lines, and if it wasn't exposed to horizontal lines soon enough, it never did learn to see horizontals well if at all.

What this amounts to is that shaping behaviour is equivalent to shaping physiology, a principle that is too general to be of much use. Tracing the pathways of nerve impulses from retina to VC should in principle indicate how the illusion is generated, but, as your outline indicates, it ain't easy. The fact that people (me) can learn to control some illusions suggests that those pathways will be more or less subtly changed by the mere act of seeing the illusion, for one thing.

I really can see no way to leave behaviour out of it.

HTH
.

Relevant Pages

  • Re: Cool visual illusion
    ... >> Wolf Kirchmeir wrote: ... >>>Except that knowledge of the illusion IS knowledge about the internal ... several classes of bipolar and horizontal cells at the ... whatever happens at higher levels in the brain. ...
    (comp.ai.philosophy)
  • Re: Cool visual illusion
    ... "Except that internal mechanisms are knowledge of the illusion." ... several classes of bipolar and horizontal cells at the ... and that's just the retina. ...
    (comp.ai.philosophy)
  • Re: Cool visual illusion
    ... > Except that knowledge of the illusion IS knowledge about the internal ... several classes of bipolar and horizontal cells at the ... and that's just the retina. ... whatever happens at higher levels in the brain. ...
    (comp.ai.philosophy)
  • Re: Cool visual illusion
    ... > would you insist that what the brain does is not behaviour? ... > the illusion available while you're doing those measurements. ... but finally most these cells' responses are filtered out. ... > VC generates occur in very small clusters of cells, ...
    (comp.ai.philosophy)
  • Re: Cool visual illusion
    ... several classes of bipolar and horizontal cells at the ... whatever happens at higher levels in the brain. ... > knowledge about the internal mechanisms ...", ... If they cause an external effect, then you can understand them by studying ...
    (comp.ai.philosophy)