What selects behavior?

[From Bill Powers (931002.1130 MDT)]

Avery Andrews (9210021.0930) --

As for termite nests vs. crowd-rings, I think there is a
significant difference between the two cases, which is that
there is probably no selective pressure on people to produce
crowd-rings, but there is a lot of selective pressure on
termites to produce nests, even tho they don't know that
they're doing this.

I think it's important to recognize that whether we're talking
about the behavior of people in groups or about termites, we're
using our own perceptual interpretations to describe what's going
on. The people in a crowd do not form arcs and rings; termites do
not built nests. In either case, the individuals are controlling
perceptions in such a way that arcs, rings, or nests appear, but
the perceptions under control have nothing to do with controlling
the aspects of the environment that make us think of arcs, rings,
and nests. Arcs, rings, and nests are effects on human
perceptions created by the actions of the behaving systems: side-
effects. Simply by looking at these side-effects, we can't tell
what's important about them, if anything.

A termites' nest, in some parts of the world, may take the form
of a dirt mound five feet high and two or three feet in diameter.
Does this mean that natural selection has selected for building a
nest with these dimensions, or for the fact that the nest is
built above ground? Not at all; in fact, those aspects of the
nest are probably of no significance to any termite. To
understand why the nest appears as it does, we would have to
discover what the actual survival factors are that are affected
by the actions we call "nest-building" (which are not nest-
building at all, from the termites' point of view). We would have
to get inside a termite and see how it feels to be inside a nest
-- what the temperature is, the humidity, the path length to
various places where critical things happen like the laying of
eggs and feeding of young, and so forth. Natural selection will
operate to preserve just exactly those variables controlled by
the individual termites that bear directly on reproduction. All
other effects of the termites' behavior, however organized or
interesting they may look to us, are irrelevant.

Natural selection is not selecting for environmental conditions
but for conditions inside the termites. What is being selected is
a set of reference signals and the control systems to go with
them. The nest that ultimately appears in human perceptions is a
by-product of each individual's controlling particular variables
at particular reference levels. Among all the uncountable
byproducts, a few affect the termites in ways of importance to
reproduction. What evolution has taught the termites is what
variables need to be controlled in order that certain byproducts
of the control process, unknown to the termites and to us, remain
in states that promote successful reproduction. Those byproducts
may have nothing to do with the ones we notice.

Suppose we try to set up the CROWD program so it will evolve to
produce arcs and rings. What would we use as the survival
criteria that would lead to this result? If we followed the naive
biological "genetic algorithm" approach, we would make survival
depend on the formation of arcs and rings. Individuals whose
behavior promoted the formation of arcs and rings would survive;
the rest would be weeded out. Eventually we would be left with
only those individuals whose behavior promoted the formation of
arcs and rings.

That might eventually work, but does it make organismic sense? We
have imposed from outside the organism a condition that makes
sense in our own perceptions, but has nothing to do with the
perceptions or experiences of the individuals in the program.

In fact there is a much more "organism-like" approach that has
nothing to do with arcs and rings. All we have to do is make food
available to the leader at the destination, provide that the
others get food if they stay close but not too close to the
leader, and provide penalties for collisions. This would
eventually result in the acquisition of the appropriate control
systems by the individuals: destination-seeking, person-seeking,
and collision-avoidance. Those would be the control systems that
actually promote survival by putting each individual into the
position it needs to be in to survive.

The outcome of having these control systems is what we see: the
leader winds a path through obstacles to the destination, the
followers follow the leader while avoiding obstacles, and when
the leader stops the followers form an arc or a ring around the
leader. But these visual patterns in the eye of the observer have
nothing to do either with the controlled variables and control
systems in each individual or with the evolutionary reasons for
controlling those variables. The survival criteria are those that
matter to each organism, not to the observer.

In general, the organized patterns we see in the behavior of
organisms are not very likely to be relevant to that behavior --
either to the behavior that produces the patterns, or to the
evolutionary forces that shaped the organism to produce the
behavior that results in the patterns. We need a Test for the
Evolved Control System just as much as the Test for the
Controlled Variable. Without some systematic way of finding out
just what survival factor is actually involved, we will continue
to think, naively, that all we need to do is to look and we will
see the survival factor that accounts for the behavior.



Bill P.