[Martin Taylor 941003 17:45]
Bill Powers (940928.0850 MDT)
On the proposed "nutrient bath" evolution simulator:
Would it not be better to have the "selection-control" variable be
internal to the organism, rather than a sensed environmental quantity?Well, sensing the concentration of nutrients would generate a variable
inside the organism, wouldn't it? I can imagine messenger molecules that
represent that concentration, but it's harder to think of how "rate of
growth" could be sensed. Why not just sense the variable on which rate
of growth depends, the amount of nutrient crossing the cell wall? But
your idea can certainly be tried.What about Gactual-Gmin (actual growth rate minus growth rate below
which death occurs)?That seems more external to the operation of the system than sensing
nutrient concentrations would be.
Yes, you are right. I was objecting to the idea that an external variable
would control bifurcation and death, but my substitution was worse. If
we consider the metaphor of ingestion and excretion, perhaps the entities
could excrete at a rate linearly related to a difference between an
internal and the external concentration, and ingest at a rate proportional
to some other function of size. About what function, see below.
One idea I had was that the rate of absorption of
nutrients, beside depending on nutrient concentration, should depend on
surface area. Growth itself increases volume faster than surface area.
I'd have to get something in there about metabolic requirements which
increase with volume, so the rate of growth in a constant nutritive bath
would go as 1/radius. Bigger organisms would have a harder time
ingesting enough to sustain growth and metabolism.
This is the "reason" (thinking of the metaphor of evolution as designer)
that larger organisms have relatively long tubes through which to absorb
nutrients, whereas very small ones may even be nearly spherical. I think
you could imagine your evolving organisms to obtain nutrients a little
more readily than the 1/r function might suggest. Perhaps 1/sqrt(r), or
something--or it might be another of those genetically controlled variables
in a far future incarnation of the program. (I imagine that there is a
well-known relation between nutrient ingention surface area and body vlume
in real organisms--there is for just about anything else, and the relation
is usually some kind of power law).
The reason for mentioning this is that to place a 1/r penalty on growth
is likely to do strange things to the fitness, and to require quite a bit
of hand-tuning of parameters you would like to allow to find their own
values. Wrong choices here could lead to whole populations dying before
they get properly started.
Remember that I'm programming in C, which is quite a lot faster than
Hypercard. I get a thousand or so reproductions per second, even with
cleaning out the casualties after every time interval.
I'm not sure how to make a comparison. It would be possible for me to
take your code and turn it into a Hypercard XCMD, provided that the
graphical part is carefully segregated and no graphic-related assumptions
are implicit in the evolutionary part of the code. Comparing that with
the same algorithm running directly in HyperTalk would be the best comparison,
because it would be free of problems of machine speed and the like. I
think HyperCard is slow, because in my experiments it takes about 3 seconds
per generation of 200 organisms. Your thousand per second is about 15 times
as fast, if you were doing essentially the same work. If I thought that
this could all be attributed to Hypertalk versus C, I'd think about putting
a bit of effort into redoing things in C instead (I have both C++ and Pascal
at home, but it's a long time since I did any serious programming; I use
Hypertalk because it is easy to deal with, especially for displaying this
kind of stuff).
···
=================================
Bill Powers (940930.0655 MDT)
Martin Taylor (940929.1030)--
... access to an IBM-compatible PCT (Personal Computer Terminal). That
is not always the case.The Mac is a wonderful machine for the user. But behind those simple
icons, windows, dialogue boxes, sliders, etc. is a lot of very complex
programming....
It would be great if some smart, generous, young Mac Whiz were to help
out ...Until something like that happens, my programs will be written for PCs,
not because PCs are better than Macs but because programming them is
within my capabilities.
My comments should not be construed as a request for you or anyone else
to start programming for the Mac. They were intended as a comment on
Dag's notion that there's no excuse for anyone not to have run all the
demos. I'm sure I'm not the only one who has sporadic or no access to PCs.
=============
It strikes me that these phenomena could easily be made part of a PCT
experiment. Suppose the subject saw some shaded bumps and hollows on the
screen, with the apparent depth being variable and affectable by a
handle and a disturbance. Every time the depth went through zero there
would be a chance for an ambiguity (and maybe even when it is not zero).
I'm not sure what you propose. Subjects don't perceive the dspth of the
bumps and hollows "going through" zero. They perceive that there are
bumps or that there are dents. Now if what you propose is that the shading
should be affected by the subject's handle, I don't think that would work,
because the same shading applies both ways, so the same movement of the
handle would increase the apparent height of a bump or the apparent depth
of a dent. There wouldn't be any reversal of handle-height effect. Now,
if you want the subjects to maintain always a "dent" perception, by
reversing the apparent direction of the light, and using the HANDLE to
make the apparent depth go through zero, that's a different matter. The
physical pattern would be changing, and that would change the nature of
the random walk of the (presumed) lower-level perceptual units that
is theorized to occur when they are exposed to an invariant physical
pattern. The experiment might work, but it would be dealing with something
different.
Would putting such figures in a control loop make the flips less random?
I have supposed so, in the past, but have never tried it. My reason for
supposing so is that it is through the effects of one's actions on one's
perception that one acquires a sense of the solidity and reliability of
the world. If one is reliably controlling some aspect of an otherwise
ambiguous perception, it seems to me likely that the perception will be
more stable. But that's just an intuition.
==================
The same posting of Bill's has interesting comments on replication in
science, on which I would like to comment, but have no time now. Maybe
later. Basically, I think "triangulation" is worth a lot more than
"replication." If a perception can be controlled with input from either
of two quite different sources, it is more likely to represent some
consistency of the world than if it continues to be controllable with
input from always the same single source. Science is a matter of
formalizing the control of perception, and the same should apply. I put
little store in replication, but much more in perceptual constructs that
reappear when the subject domain is addressed in a variety of ways.
And I'm not justifying all that body of psychology and sociology that
correctly receives such bad press here.
Martin