Simon and ants

[From Bill Powers (990425.0636 MDT)]

Bruce Abbott (990423) --

Unlike most, I have actually _read_ Simon's "Sciences of the Artificial."
What Simon was talking about was that the ant, as a behaving organism in
this situation, is basically simple. The complexity seen in its behavior is
due, not to the complexity of the ant, but to the complexity of the
environment.

Given that the ant's behavior in the situation described (returning to its
nest over rough, obstacle-filled terrain) can be elegantly described by the
few simple equations that specify a particular control system, I find
Simon's observation right on the money. The complexity of the ant's
behavior is due not to the complexity of the ant but to the many
disturbances which must be countered, that threaten to deviate the ant's
path from its reference direction.

That's all very fine as a retrospective interpretation, but Simon had no
such explanation of the ant's behavior. He thought that the complexity of
the ant's behavior was caused by _stimulation_ from the environment, and by
forces deviating the ant from its path. That's why he referred to the
organism as a giant input-output function. He had observations of the
actual behavior of ants
as a constraint, as do we. His explanation naturally was consistent with
the observations. But it was wrong.

The true explanation, I claim, lies in HPCT. The crowd program shows how.
The complexity of an individual's path in the crowd program does not come
from "disturbances", but from possessing multiple goals: to avoid
collisions, to remain at a specific distance and angle from another person,
to increase the proximity to a goal-position. These multiple internal
goals, plus the layout of the obstacles and the location of the
destination, explain the path that is taken. The environment exerts no
actual disturbing forces on the individual other than gravity, nor is the
individual optimizing anything.

Simon played around with feedback concepts and goal-oriented behavior for
his whole professional life. But he emphatically rejected the idea that
control theory was involved. I know, because he was among the famous people
I wrote to back in the beginning before I found out that famous people are
very competitive -- that's how they get famous. He said that control theory
had nothing to do with the goals in his Logic Theorist, or his Theorem
Prover or chess-playing programs. He seemed angry that I would suggest such
a thing.

Best,

Bill P.

[From Bruce Abbott (990425.1600 EST)]

Bill Powers (990425.0636 MDT) --

Bruce Abbott (990423)

Given that the ant's behavior in the situation described (returning to its
nest over rough, obstacle-filled terrain) can be elegantly described by the
few simple equations that specify a particular control system, I find
Simon's observation right on the money. The complexity of the ant's
behavior is due not to the complexity of the ant but to the many
disturbances which must be countered, that threaten to deviate the ant's
path from its reference direction.

That's all very fine as a retrospective interpretation, but Simon had no
such explanation of the ant's behavior.

It's my retrospective interpretation, not Simon's, that's true. But what
Simon actually did say -- that compexity of the ant's path toward its nest
is due to the complexity of its environment rather than to any necessary
complexity of the ant -- is certainly true in the sense that the ant-side of
the equations that describe the ant-environment system need not be very
complicated.

The true explanation, I claim, lies in HPCT. The crowd program shows how.
The complexity of an individual's path in the crowd program does not come
from "disturbances", but from possessing multiple goals: to avoid
collisions, to remain at a specific distance and angle from another person,
to increase the proximity to a goal-position. These multiple internal
goals, plus the layout of the obstacles and the location of the
destination, explain the path that is taken. The environment exerts no
actual disturbing forces on the individual other than gravity, nor is the
individual optimizing anything.

I agree that it would take more than a simple "elementary control unit" to
properly model the ant, even in this restricted situation where we are only
concerned with accounting for its path back to the home nest. But to say
that the complexity of the path does not come from "disturbances" is
stretching mightily to make a point, Bill. If the ant just kept walking on
a true heading toward the nest, it would soon encounter barriers and
deflecting surfaces that would either push it off course or bring it to a
complete halt. In their absence the ant would make a bee-line (aunt-line?)
for the nest. Thus, bringing in additional control systems with their
sometimes conflicting requirements doesn't change the fact that the complex
path observed is a consequence of the complexity of the environment and not
the complexity of the ant. In the absence of obstacles the complexity of the
ant to which you refer still remains, yet the path is simple.

Bruce

[From Bruce Gregory (990425.2100 EDT)]

Bruce Abbott (990425.1600 EST)

I agree that it would take more than a simple "elementary control unit" to
properly model the ant, even in this restricted situation where
we are only
concerned with accounting for its path back to the home nest. But to say
that the complexity of the path does not come from "disturbances" is
stretching mightily to make a point, Bill. If the ant just kept
walking on
a true heading toward the nest, it would soon encounter barriers and
deflecting surfaces that would either push it off course or bring it to a
complete halt. In their absence the ant would make a bee-line
(aunt-line?)
for the nest. Thus, bringing in additional control systems with their
sometimes conflicting requirements doesn't change the fact that
the complex
path observed is a consequence of the complexity of the
environment and not
the complexity of the ant. In the absence of obstacles the
complexity of the
ant to which you refer still remains, yet the path is simple.

I think Bill's point is that one of the ant's goals is not to collide with
obstacles. So that taking the path around the obstacle is the result of the
ant's goal, not the existence of the obstacle. Granted that the path would
be more complex with more obstacles, but this complexity would still be the
result of the existence of the ant's goals.

Bruce Gregory

[From Bill Powers (990426.0631 MDT)]

Bruce Abbott (990425.1600 EST)]

But what
Simon actually did say -- that compexity of the ant's path toward its nest
is due to the complexity of its environment rather than to any necessary
complexity of the ant -- is certainly true in the sense that the ant-side of
the equations that describe the ant-environment system need not be very
complicated.

When you have modeled an ant, you will have earned the right to say how
complex it is.

Simon's point, frequently made, was that the actions of the behaving system
do not need to appear as complex as the details of its interactions with
the environment appear to a naive observer. If Simon had spoken in terms of
appearances and naive observers, I would agree with that. However, the
interactions with the environment are primarily a consequence of the ant's
pursuit of its own goals. On a sandy beach an ant's movements can be
extremely complex as it maneuvers among the grain but the grains do not
initiate its movements. The environment, in terms of its ability to produce
intended results, is far less complex than the ant. Don't confuse the
presence of a lot of geometric detail with behavioral complexity.

I agree that it would take more than a simple "elementary control unit" to
properly model the ant, even in this restricted situation where we are only
concerned with accounting for its path back to the home nest.

If you are recording data about the ant only to the extent necessary to
observe that is it moving toward its nest, then you should be doing the
same for its environment. If you are observing the shifting of every grain
of sand touched by the ant, then you should also be observing the
infinitely detailed adjustments of the ant's muscles that prevent the
shifts from affecting what the ant is controlling. If you are including
possible odors in the air and on the ground, and the ant's possible control
of them by altering its path temporarily, you might begin to understand the
complexities of the ant's behavior.

But to say
that the complexity of the path does not come from "disturbances" is
stretching mightily to make a point, Bill. If the ant just kept walking on
a true heading toward the nest, it would soon encounter barriers and
deflecting surfaces that would either push it off course or bring it to a
complete halt.

The disturbances of which you speak are caused by the ant, not by its
environment. If the ant were not moving, the lumps of dirt, twigs, grains
of sand, and so on would not seek out the ant and exert forces on it. I
agree that the ant's encounters with obstacles (for example, climbing over
them) do result in changing forces on the ant, but since those forces are a
function of the ant's own output actions, they are consequences of the
ant's behavior, not the behavior of its environment. They are caused by the
ant's qo acting via the environmental feedback function Fe on qi.
Disturbances are external influences, qd, acting through a disturbance
function Fd on qi independently of the effects of qo.

In a somewhat more complex model we would include _parametric_
disturbances, which is to say, shifts in the form of the environmental
feedback function caused by independent influences. In this case, however,
the shift in the form of that function is again produced by the ant's own
output. When a grain of sand rolls as the ant's foot comes down on it, we
find that the feedback function is more complex than we had initially given
it credit for. But the ant's adjustment to this shift is equally complex,
if it prevents the shifting grain from altering the ant's orientation or
movement.

In their absence the ant would make a bee-line (aunt-line?)
for the nest.

You are assuming that the _only_ goal is to return to the nest. I doubt
that this is ever the case.

Thus, bringing in additional control systems with their
sometimes conflicting requirements doesn't change the fact that the complex
path observed is a consequence of the complexity of the environment and not
the complexity of the ant. In the absence of obstacles the complexity of the
ant to which you refer still remains, yet the path is simple.

You are now asserting that the complexity of the ant can be greater than
that of its environment. In fact, if the ant is successfully counteracting
all effects of environmental variations on all of its controlled variables,
then the complexity of its _actions_ exactly matches the complexity of
environmental disturbances, with enough variation left over to accomplish
the ant's purposes.
In a simple (disturbance free) environment, the actions are simple; in a
complex environment, they are complex.

Also, you are making up an observation to fit the conclusion you want to
draw. I think it would be very hard to find an ant that simply headed in a
straight line for its nest, even in the total absence of irregularities in
the terrain. How do you know that the ant isn't following scent trails of
other ants (and itself) laid down when it left the nest? When ants leave a
nest they move in all sorts of irregular-seeming directions. If they're
following scent trails back, they will also move in irregular-seeming
directions, even though their behavior is perfectly systematic and there
are no obstacles.

Mary and Tom Bourbon and I once spent a couple of hours watching individual
ants at a campground. It was extremely hard to identify any disturbances,
or any controlled variables. The behavior of the ants seemed to be
independent of anything we could identify just by looking. We decided that
we didn't understand what ants control for.

Simon's statements about ants were glib, not scientifically useful.

Best,

Bill P.

[From Bruce Abbott (990426.1435 EST)]

Bruce Gregory (990425.2100 EDT) --

I think Bill's point is that one of the ant's goals is not to collide with
obstacles. So that taking the path around the obstacle is the result of the
ant's goal, not the existence of the obstacle. Granted that the path would
be more complex with more obstacles, but this complexity would still be the
result of the existence of the ant's goals.

Thanks, Bruce. It's a good point (and one I understand), but it doesn't
change the conclusion. The complexity of the ant's path is still due to the
complexity of the terrain through which the ant must move, not to the
complexity of the ant.

Whether the ant is forced to take a circuitous path because it runs headlong
into a ridge of sand and is physically deflected from its course, or because
it senses that it is encountering a ridge of sand and thus must alter its
heading in order to maintain a certain distance from the ridge, the
deflection of course arises because the ridge is in the way. The ridge
doesn't physically disturb the ant's heading, but the actions initiated by
the object-avoidance system disturb the heading just as surely. Similarly,
it is the forward motion of the ant (another control system responsible for
that) that brings the ant into such proximity to the ridge that its
proximity-control system is disturbed and takes action to change the ant's
course.

Regards,

Bruce A.