[From Bill Powers (941226.0900 MST)]
Bruce Abbott (941220.1400 EST)--
Don't know what happened with that program I sent you -- maybe I had the
wrong format in effect for creating ascii files from my word processor.
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?
I see you got around my problem with using reorganization at the second
level. In order to keep the organism from dying, I found I had to make
the reorganizing process very slow -- but that allowed the organism to
get into regions where the gradient was so low that everything slowed to
a stop. I didn't keep on with the project, so don't know if I could have
found a solution.
Yours is simply to start E. coli over again each time it "dies," with a
new set of parameters and a new supply of stored nutrient. I notice that
the new organism is also set back onto the playing field if its ancestor
had gone off it. So basically your model keeps restarting with new
parameters but from similar initial conditions until a set of parameters
is found that results in a control system.
A problem with realism is that each new E. coli is given a new store of
nutrients, which would not be the case in the real E. coli. Nor,
considering the method of reproduction, would the mutated offspring be
relocated into the region of normal concentrations.
Your evolutionary model could be made into a reorganization model if you
didn't require E. coli to die. By setting a nonzero lower level of
stored nutrient as the trigger, and randomly shuffling the parameters
each time, you might be able to make E. coli survive, while still
reorganizing. You'd need to use rate of change of stored nutrient, too,
so that if the level of stored nutrient started to rise while the
absolute level was still too low, you wouldn't get another
reorganization right way.
You could make the model even more interesting if you allowed E. coli to
reproduce instead of "bursting" at a nutrient level of 200.
ยทยทยท
---------------------------------------
I still don't see any selection by consequences, except in a sort of
roundabout metaphorical sense. The consequence of dying doesn't really
select anything, does it? As far as I can see, when an organism dies
that is the end of it. Its death doesn't affect anything else. Of course
in your program the routine for creating a new organism is triggered by
the death of an old one, and doesn't execute unless an organism dies. So
we could say that creating a new organism is "selected" by the death of
an old one. But inside that routine, I can't see any selection process
-- you just get random new parameters, plus a refill on the stored
nutrient and a boost back to a reasonable place in the gradient. The
chances are probably appreciable that the new organism "selected" in
this way will die, too (although the number of deaths seems remarkably
small). This continues until no more deaths occur.
The problem with using a word like selection, and saying that the
consequence is doing it, is that I would expect to see some selection
mechanism at work in the thing that is said to do the selecting. I can
see that the overall effect is AS IF the consequence could select a
behavior that produces it, but I always assume that we're speaking
literally, particularly when proposing models. In a control-system
model, the way the system behaves when the reference signal is constant
suggests that the disturbance directly causes the output of the system
that opposes the disturbance. But that is an appearance only, and not
the literal truth. Likewise, it seems that the reference signal directly
sets the value of the input quantity, but again that is only an
appearance. When we understand how the control mechanism actually works,
we can see how these appearances are created, but we no longer believe
that the appearances are telling the true story.
In the crowd program, when you see an active individual wend its way
through a maze of obstacles to reach a goal-circle, it really looks as
if the goal-circle is selecting behaviors by the individual that will
eventually bring it inside the circle. Someone who didn't know about the
model behind the Crowd program, and who was unfettered by prejudices
against magic, might actually propose some such model, saying that the
goal-circle was an attractor of some sort, pulling on the individual and
making it move toward the source of attraction. Heck, look at Kurt
Lewin's "field theory" of the 1930s, which actually proposed that
objects in the environment attracted and repelled organisms in their
vicinity. I don't know if even Lewin took that theory literally, but he
seemed to talk as if it were literally true.
I think that metaphorical descriptions of behavior create more confusion
than understanding. "Goal-seeking" behavior is often talked about as if
the goal somehow exists in the environment, so that even if it doesn't
actually exist, it is still sought by the organism. At the time I say
that my goal is to be at the gas station at the foot of Farmington Hill,
my actual position is my front driveway. I am NOT at the gas station. So
how can my being at the gas station, a condition that does not exist,
result in my being at the gas station? Without the PCT model, this
problem just becomes a big muddle, and the whole muddle is caused by
using a metaphor that implies a nonsensical mechanism. In fact,
organisms do not "seek goals" if by "goal" you mean some condition in
the environment that does not actually exist. What they seek is to make
perceptions of the actual environment match internal reference signals,
which DO have physical existence. When you understand that, and adjust
your language (or at least your meanings) accordingly, the muddle
disappears.
So while I can understand "selection by consequences" in a metaphorical
sense, I can't accept it as a literal description. The mechanisms behind
this process that we get out of PCT don't contain any selection devices
or procedures; there is something else entirely going on. We may
describe the appearances created by these mechanisms, poetically, as
"selection," but anyone taking that description literally would be hard
put to point to the selector -- and the means by which it selects.
-----------------------------------------------------------------------
Best,
Bill P.