[From Bruce Abbott (941227.10 EST)]
If I can believe my e-mail this morning, we are back on line. Looks like I
have a LOT of catching up to do.
Rick Marken (941222.1630)
I just don't understand how you think your evolution model is a
reorganization model? I need a diagram to see it. Did you see my
diagram of the reorganization system? Maybe you could draw an analogous
diagram of your reorganization system. Then whatever point you are trying to
make would be clearer to those of us who are despretly trying to interpret
you comments as reflecting a lack of understanding of PCT.
Bill Powers (941226.0900 MST)
Your new "selection" program is very neatly constructed and works just
as you say. I looked it over fairly carefully, however, and couldn't
find the part of the code that is doing the selecting. Maybe I just
didn't recognize it. Can you point it out to me?
Rick Marken (941226.1745)]
Could you please, Bruce? This is why I asked for the diagram of the
control loop in your "selection" program. I just didn't see what might
be doing the "selecting". You can imagine how important such
clarification must be for me if even Bill needs it;-)
O.K., let's take it from the top. When I posted that model, I never said
anything about its being a reorganization model. Take a look at the post.
What I DID say was that it was an illustration of selection by consequences
that even Rick Marken would like. It was only after Rick responded to that
post asking whether I intended it as a reorganization model that I began to
explore whether the model might indeed serve that function.
In the demo there is random variation in control system parameters; those
control systems that fail to adequately control the level of stored nutrients
cease to exist and are replaced by systems with yet other parameters.
Eventually you get a set of four bugs that "function" in this environment.
What could be simpler? Here's the diagram you requested:
random
+-->parameter -------+
> variation |
> V
<----death<--- Nutrient Control
System
^ |
> V
Environment
Note that there is feedback in this system: death of an organism ("error")
leads to the selection of new parameters (for a new organism, of course); the
process continues until a set of parameters is found that works.
If for "death," we substitute some other suitable criterion (e.g., persistent
or extreme error in the nutrient control system), we have a system that
reorganizes itself until it (a) dies or (b) finds a set of parameters that
control. Thus the evolution model does provide a framework for constructing a
reorganizing system. Whether such a system would lead to a viable set of
parameters in time is another matter.
Rick, if you've followed me thus far, you are now in a better position to make
sense of our previous exchange:
Me:
Are you saying that this new demo shows that control (purposeful behavior)
can result from selection by consequences?
Bruce:
Yes, of course!
Me:
If so, what is the controlled variable?
Bruce:
Stored nutrient level ("fuel").
I am referring to the fact that the process of random variation and natural
selection eventually finds control systems that function to control the level
of stored nutrients. For some reason, you seem to have arrived at the belief
that this is demo contains what I believed to be a model of a reorganizing
system that would act on an individual e. coli rather than on the species. It
doesn't, and I never claimed that it does.
Now, as to the phrase "selection by consequences." The "consequences"
(survival or death) obviously do not select the parameters in any literal
sense: these are selected entirely at random. Yet in another sense they do.
Death causes another round of parameter selection to occur; death causes
reselection, survival prevents it. I mean "selection by consequences" in the
latter sense: that the consequences drive the selection process, not that they
do the actual selecting. Rick seems to be using this phrase in the same sense
in the following:
Rick Marken (941220.1745)
You seem to think that I don't believe in "selection BY consequences".
But I do. I think your Ecoli4 program used "selection BY consequences" to
change the probabilities of tumbling. I think sand filtering through a seive
is "selection BY consequences"; the large grains are "selected" to stay in
the seive by a consequence of their behavior; as they move down (courtesy of
gravity) they push against (rather than through) the wire mesh.
"Selection BY consequences" exists BUT IT DOES NOT PRODUCE CONTROL.
And, to the extent that consequences select the outputs (or the behavior)
of a system that is already organized as a control system (as was your
E. coli model) they INTERFERE with the system's ability to control.
I agree that the evolutionary model you posted (assuming that it
works; I still have no Turbo for the PC) operates on the basis of
"selection BY consequences". I would imagine that there will be a
bunch of bugs with "good parameters" running around after a few
selection episodes.
The essential fact here is that the consequences (in this case survival/death)
determine whether or not there will be a new round of (random) selection. In
this sense they act as criteria by which the program selects whether or not to
generate a new set of control parameters. The phrase "selection OF
consequences" fails to call one's attention to the criteria upon which the
selection is based, even though it is clear that the program is doing the
selecting. Given that the consequences, as criteria, determine what decision
the program will make, one can fairly say that selection is determined by
consequences. But the loop is closed. Selection determines consequences
which determines whether or not there will be a new round of selection. Thus
both interpretations are correct.
Regards,
Bruce