[From Bill Powers (941227.1145 MST)]
Bruce Abbott (941227.1000 EST) --
Selection determines consequences which determines whether or not there
will be a new round of selection. Thus both interpretations are
correct.
This is probably the core of our semantic difficulties. In your program,
behavior determines a consequence, which happens in this case to be
death. This consequence, however, does not determine anything without
the help of another part of the model: the part that detects death and
takes it as the occasion for creating a new organism. The behavior of
the new organism (in the given environment) then determines whether
death will occur or not, and so forth. This active system is designed to
replace dead organisms with live ones, and clearly it (or its
programmer) prefers live ones.
To say that consequences select behavior is to leave out the part played
by an active agency that detects the consequences, compares them with
specifications for a desired consequence, and changes the behavior or
the behaving system accordingly. When you include that active agency,
you find that the selection is being done not by the consequences but by
some system that is evaluating the consequences.
Darwin named "natural selection" to be analogous to purposive selection
as practiced by animal breeders. Purposive selection is easy to
understand: the breeder crosses the strains of pigeons, looks at the
consequences in the form of many random variants on pigeonness, and
picks the variant that is desired for further propagation. So it is
clearly the breeder who has a reference criterion for selection, and it
is this criterion that determines which pigeons will be further
propagated. It is clearly not the offspring which are the consequences
of breeding that select pigeons which are to propagate: it is the
breeder with an internal process of perception and comparison who
carries out this selection.
Natural selection is not analogous to purposive selection. The only
process in common is the random variation in the consequence-producing
behavior (and, trivially, the word "selection"). In purposive selection,
the consequences are weighed in relation to a desired consequence, and
if there is a difference, another round of random crossings is
instituted. If the difference is lessened, then those invididuals
showing characteristics closer to the desired ones are selected for
further breeding. When the difference is zero, cross-breeding is
replaced by inbreeding, to stop the random changes.
In natural selection, there is no active agency that desires any
particular consequence to occur. There is, in fact, no selection process
at all. The consequences are whatever they are, and there is no change
in the mutation process that would favor any consequence over any other:
by classical evolutionary theory, all consequences remain equiprobable.
It is only by analogy that we say that nature selects the fittest, as if
nature had some preference to see more fit organisms in it. Because some
organisms fail to survive, we say that the remaining ones were
"selected." By implication, we say that nature must have had some
preference for the survivors.
But only living organisms have preferences; the blind forces of nature
have only consequences and have no preference for one consequence over
another. So the analogy of natural selection to purposive selection is
flawed.
As workers in biochemistry and cellular biology have been discovering
lately, purposive selection is far more effective than natural selection
could possibly be. A soup of DNA is chopped up into random fragments and
then the experimenter selects random recombinations that fit his
criterion for selection. In this way new forms can be very quickly
achieved. As soon as a goal is involved, selection guided by perception
and comparison with a reference standard becomes very efficient.
I think Stephen J. Gould was on the right track when he said that the
record of evolution should be viewed, theoretically, as a bush, not a
tree. He insists that there is no direction in evolution, not even
toward increased fitness. To say that the unfit fail to propagate is not
to say that the fit are any more fit than the survivors of many
generations ago were. By following the logic of classical evolution as
closely and literally as possible, we must conclude that evolution peels
away the unfit but in no way leads to an increase in fitness. Gould was
simply being as faithful as possible to the core concept of basic
evolutionary theory.
I happen to think that Gould was correct in his interpretation of basic
evolutionary theory, but that what he has shown is that basic
evolutionary theory is inadequate. While natural selection may account
for the very first stages of life, I don't think it can carry us much
further than that. At some point, natural selection must have produced
systems capable of purposive selection (of which our pigeon breeders,
marriage brokers, and arrangers of debutant parties are recent
examples). This does not mean what religion-shy evolutionary theorists
will immediately assume; it does not mean some extra-physical external
agency seeing to it that evolution proceeds in some predetermined
fashion. It means only that organisms must have evolved quite early so
as to be able to select their own forms according to criteria carried in
the organisms.
Of course what we humans see as "forms" is quite irrelevant to what
organisms would be concerned with at the evolutionary level. Organisms
or species can't care how big they are, or how many appendages and
sensors and muscles or cilia they have, or what markings they carry.
They can't care even about their behavior at this very low level. All
those things have to be variable in order for evolution to work. What
they can care about is maintaining certain basic variables at specific
inherited reference levels, as we humans care about keeping blood CO2
below a certain level, circulating glucose above some level, and
brainstem blood temperature at some level. More generally, what they can
be concerned with is controlling what happens to themselves.
What I am suggesting is that some time very early in the evolution of
life, the principle of blind variation and _purposive_ selection
evolved. Purposive selection entails detecting some variable, comparing
it with a reference level, and instituting a blind variation in a
particular relation to the difference: sooner for large differences, and
later for small ones. The mechanism for doing this is called "mutation,"
and is generally attributed to cosmic rays or thermal noise; I am
attributing it to a capacity that organisms have evolved. Organisms that
have remained the same for several hundred million years (like that tree
they've found in Australia, or coelocanths, or cockroaches) have not
become immune to cosmic rays; they have attained a state in which the
perceived values of fundamental variables match their reference states,
so that no mutation is called for.
Biologists have been in a tizzy lately over the Cairnes et. al.
discovery of apparently directed mutations in E. Coli and other
microorganisms. They have studied the mutations very carefully, and as
far as they can tell the mutations are still random. But they miss the
point (as, apparently, do Cairnes et. al., as they have not yet offered
the argument that I offer). Selection occurs at the input, not the
output. It is not necessary for the mutations themselves to be
nonrandom. What is necessary is for a selection system to exist in E.
coli that can produce mutations more frequently when the system is
stressed, and less frequently when it is not. This alone is sufficient
to bias the _consequences_ of even purely random mutations in a
favorable direction. But of course this implies that E. Coli contains,
at a very basic and inheritable biochemical level, sensors that can
report specific consequences of mutations, and reference signals that
define what a desirable state of each consequence would be. In other
words, E. coli and other microorganisms that show the "SOS" or other
mutational responses to stress must contain control systems organized in
the same way that E. coli's method of locomotion is organized.
E. coli demonstrates purposive selection from random consequences of its
behavior. It contains a reference signal defining the minimum acceptable
level of the rate of change of sensed concentration of an attractant, or
the maximum acceptable level for a repellent. This reference level,
combined with the right output function, results in varying the timing
of tumbles without in any way affecting their randomness, and in so
doing steers E. coli up or down the gradient almost as efficiently as if
it could vary its direction of swimming systematically.
So E. coli provides us with a proof-of-principle, the principle of blind
variation and purposive selection. We now know that it works, it works
efficiently, and it works by mechanisms that we can understand. Since we
now know that this principle works, we can begin to consider that it
might work at even deeper levels of functioning, down to the level where
the random changes in output are what we call mutations.
The natural result of random variation and purposive selection is a
continual increase in the ability of an organism to resist disturbances
and determine the states of environmental variables that have important
effects on it. In short, continued evolution of this kind leads to
better and better control, as long as better control is required. So
there is a directional arrow behind evolution: organisms, when
necessary, evolve to improve their control over the local environment.
Evolution, seen from this standpoint, is a form of behavior, and
selection is always selection BY the organism of consequences it
prefers.
ยทยทยท
----------------------------------------------------------------------
Best,
Bill P.