I am coming back with hundreds of CSGnet mails unchecked. Summer is absolutely a hot season for PCT.
I wrote an essay (attached) on evolving control systems these days with my expierences on PCT and special interest in reorganization process. Though reasoning mistakes and sentences lead to misunerstanding might exist, I hope this article could bring new perspectives and ideas to the group.
Best regards,
Bo
Evolving Control Systems_rv.doc (975 KB)
Bo Wang (2007.07.20)
[From Bill Powers (2007.07.20.0855 MDT)]
I wrote an essay
(attached) on evolving control systems these days with my expierences on
PCT and special interest in reorganization process. Though reasoning
mistakes and sentences lead to misunerstanding might exist, I hope this
article could bring new perspectives and ideas to the
group.
This is excellent work, and I expect to see much more of it from
you.
I particularly like the interpretation of “what the frog’s eye tells
the frog’s brain” (the title of a famous old paper):
**What the Frog’s Eye Tells the Frog’s Brain.**Lettvin, J.Y., Maturana, H.R., McCulloch, W.S., & Pitts, W.H.
(1959)
Proceedings of the IRE, Vol. 47, No. 11, pp. 1940-51.
These are some famous names in cybernetics.
It’s pretty clear that the frog can’t reorganize its outputs (or inputs)
enough to compensate for reversal of the retinal image. However, I’d like
to see experiments that ask if smaller reorganizations are still
possible, both at the output and the input. For E. coli reorganization to
work, there has to be some fairly smooth path between the initial and
final states, so there is always some indication of whether an error is
getting larger or smaller. The reversal of the eye may make the gap
between starting and ending organizations too large to bridge.
By the way, I found (via wikipedia) that there is a Monte Carlo method
involving random walks that looks related to E. coli reorganization. It
may be that “biased random walk reorganization” would be a more
informative name for this process. Not many people know that E. coli
steers by using a biased random walk.
[
](Random walk - Wikipedia)The evidence you offer for human ability to reverse something in the
loop is strong:
At the very
beginning, behaviors of the experimental person were almost the same with
the frog acting opposite to the desired side. A similar experiment could
be tried on yourself by looking at a mirror and trying to cut a specific
hair with scissors. A week later, the man with that prism is even able to
ride a bike just like normal people do. Though our eyeballs cannot
regenerate their way back after injured, our neural systems are much more
flexible.
The fact that these people can do many tasks normally after
reorganizing shows, I think, that it is the perceptual side that
reorganizes, not the output side. But you conclude that the input side
does not reorganize. There are apparently good argtuments for both
statements, so we need to invent some clever experiments. I could argue
that if it were the output side that reorganized, each new task would
require adaptation before it could be done. If the perceptual system
adapts to the sign reversal, the output functions can all remain as they
were before because the perceptual signals then behave
“normally.” But that’s just logical reasoning; we need some
actual demonstrations.
In your essay you say
Another question is
raised then, how many levels will this alteration influence? Though only
two levels, the altered level and the level upon it, are needed to be
changed, the reorganization will change all levels above the altered one.
It seems not an efficiency process.
This question could be stated more clearly. Do you mean that the
alteration also alters the organization of higher levels, or do you mean
that it disturbs the perceptions that (unaltered) higher levels are
controlling? I think the latter is what you mean. If so, then mere
disturbances are not enough to cause reorganization unless the higher
systems are unable to counteract the effects of the disturbances. Think
of each level reorganizing independently of any other levels, on the
basis of its own errors. (I think of “altering” a system as
meaning changing its parameters, not just disturbing its
variables).
Figure 2, showing changes in the number of viable neurons, relates to a
process called “pruning,” which means cutting branches off
bushes and trees. Plum trees, I suppose. I am told that in the baby and
infant, there are far more neural connections than in older children and
adults; reorganizing seems to be in part a process of eliminating useless
or wrong connections. David Goldstein told me about that – maybe he can
expand on this and give you some references. My reorganization
demonstrations start out with potential connections of all control
systems to all possible controlled variables. Reorganization gradually
strengthens the “good” connections and removes the
“bad” ones, not by “pruning” but by reducing the
weights close to zero, which has about the same effect.
It’s a nice paper, and I will make sure it’s seen at the CSG meeting next
week. This much could be published as an introductory study, but clearly
there is a very large area opening up here for exploring reorganization
of control systems. This might keep you busy for some years to come! I
don’t know of anyone else doing work like this, so you are a
pioneer.
Best,
Bill P.