🚀 go-pugleaf

[From Bill Powers (990429.0545 MDT)]

Bruce Abbott (990428.0800 EST)]

>> When the ant comes up against a passive
>>obstacle, the forces that result are caused entirely by the ant's muscles.
>
>I never thought otherwise.

You needed to think it through farther. That means that those forces are
not disturbances: see below.

>>Only if there is an independent active source of forces acting on the ant
>>would we identify it as the "disturbing variable" in the standard PCT
>>diagram.
>
>In my view, the actions of one control system may disturb a variable being
>controlled by another control system.

Yes, that is true. But this is not the same as the ant's foot encountering
a grain of sand that gives way beneath it,m or its antennae encountering a
pebble and being guided to prevent further contact. All that does is alter
the form of the feedback function momentarily. The control system has a
slightly higher or lower gain for the moment, but as long as the gain
remains high enough, and not too high, control simply continues. This kind
of "disturbance" is not a consequence of actions taken by some independent
active agency.

>Hypothetical example:  The ant is
>moving forward along a given path because of the actions of one set of
>control systems; if this forward motion brings the ant to an obstacle in its
>path, then a variable being controlled by another system (proximity) gets
>disturbed and action is initiated to counter the effect of that disturbance.
>The proximity system neither knows nor cares whether the obstacle was
>brought into proximity by "an independent active source of forces" or by the
>actions of another of the ant's own control systems.  It just "sees" a
>disturbance to its controlled variable and takes action.

Precisely how the Crowd program works. In the crowd program, an increase of
proximity above a reference level results in turning right or left and
lowering the reference level for forward speed. The greater the error, the
tighter the turn and the more the slowing. The speed control system is
actually subordinate to the turning control system (as well as to two other
systems).

The result of the right or left turns is to turn the person away from the
obstacle and prevent more than a very slight and momentary increase in the
obstacle proximity signal. Only in a tight concentration of obstacles does
the error increase significantly, and then the result is to turn in a tight
circle until the entrance to the cul-de-sac lies straight ahead and the
person exits. In truth, the people act more like bugs than intelligences.

We can, as you say, think of the turning control system as independent of
the forward speed control system. These are the independent active agencies
involved. They are both coordinated by the collision-avoidance system, the
person-following system, and the destination-seeking system, so they aren't
really independent. They are simply the means by which the ant as a whole
keeps or brings its various perceptions close to their reference levels.

The environment is not among the active agencies; it is entirely passive
and initiates no actions. It provides the stage on which the actions and
their consequences are played out by the ant.

>>If the ant's actions were totally and blindly automatic, the path taken by
>>the ant would reflect every obstacle and every disturbance.
>
>First, no one has asserted that the ant's actions are "totally and blindly
>automatic."  (I am assuming you mean totally open loop.)
>
>>But it does
>>not. Generally, the ant's path is _not_ deflected by obstacles; it is
>>independent of them.
>
>Not the paths of the ant's _I've_ observed!  My observations indicate that
>ants navigating on rough terrain do a _lot_ of direction-changing to go
>around or over obstacles in their paths.

True, they can't go _through_ obstacles. But if the obstacles were
determining their paths, you would see the same path every time an ant
encountered the same obstacle. Instead, the ant goes left, right, or over
the obstacle. If you were to do a real study of the ant's movements, I
believe you would find they are only slightly correlated with the presence
or placement of obstacles with respect to the ant's lateral direction of
movement. Ants do a lot of direction-changing, but when they re-visit the
same location they do not do the _same_ direction changes. Those direction
changes, I maintain, are not caused by the terrain, although there is
probably some small systematic relationship of ant to obstacle -- due
entirely to the ant's preferences and actions.
>
>>The ant might turn aside on encountering a large
>>pebble, or it might simply climb over the pebble and stay on course.
>
>You are now defining "course" two-dimensionally.  Do you think the ant does?

No, I suspect that the ant doesn't define "course" at all. I'm describing
my view of the ant. It probably maintains certain odors and perhaps blobby
light concentrations in states it prefers, by using its leg muscles. It may
not even distinguish its own orientation in the gravity field. It probably
steers in part by equalizing the intensity of odors detected by its two
antennae. Speaking of the ant as having the intention of "returning to its
nest" is a human anthropomorphism. The ant probably has no concepts of
space or geometry.

>>The
>>ant does not veer off of upslopes or turn down downslopes, as Simon
>>imagined. To the contrary; I know of ants which, on encountering a 6 x 6
>>post sunk into the ground, climb up it, eventually ending up in my house.
>>We are not seeing merely a confluence of forces, one due to the goal and
>>the rest due to the environment, as Simon imagined.
>
>Like the ant, you are veering off course, Bill.  I don't care what Simon may
>have thought the mechanism was.  I've only been defending Simon's insight
>that complex actions may reflect complexity in the environment rather than a
>complex internal organization.  You haven't convinced me that this insight
>is wrong.

If Simon didn't make his statements for the right reasons, he might as well
not have made them at all. The whole point lies in what you mean by the
weasel-word "reflects." If you mean that the complex actions reflect the
complex terrain in the way that a drop of water sliding down a rough plane
reflects, exactly, the path of least resistance, then your (and Simon's)
interpretation is wrong in my view; if you mean that the path taken
reflects the sum of all goals the organism has, with some small influences
from the terrain but by no means any determining influences, then you are
talking about something entirely different.

>This is unfair, because neither Simon nor I ever intended to attempt a model
>of any real ant.

It's not unfair because without a model you have no real knowledge.

>Simon began with a common observation:  watching an ant on
>the beach making its way back to its nest.

That is not what he observed. He observed the actions of an ant controlling
some set of perceptions of which he knew nothing, and he anthropmorphized
by assuming the ant wanted what he would have wanted in its place: to get
back to a location in geometric space which is a "nest". The ant may know
nothing either of a nest or of its spatial relationship to it. It may live
in a world of light and taste and odor, in which moving up or down
gradients of either result in changes it is genetically set up to prefer to
experience.

>He observed that the ant had
>many obstacles in its path, and often took a circuitous route as required to
>get around them and continue toward the nest.

He observed that the ant followed a path that seemed to him circuitous,
under the assumption that the ant was trying to take the shortest path to a
particular location. He dismissed the variations in the path as being
nothing but ostacle avoidance, but I doubt very much whether he carried
with him the equipment necessary to find out if the path bore any
relationship to the placements of what he saw as being obstacles. In
effect, Simon assumed that the ant was too simple to be generating a
complex series of path changes, each with a specific purpose, so he
interpreted the path changes as being caused entirely by the obstacles. And
from this assumption, he drew the "insightful" conclusion that the path
deviations from a straight line to the next were caused by obstacles. In
other words, his insight consisted of reiterating the prejudices he brought
to his observations.

>He noted that if you drew the
>ant's path on paper it would appear to be complex -- lots of loops and
>wiggles and sudden turns -- but that this complexity need not imply a
>parallel complexity within the ant.  The path might be capable of being
>fully accounted for by some fairly simple "rules" (my word) operating within
>the ant, interacting with the complexities of the terrain the ant must cross.

That is a possibility. But another possibility is that some unknown
fraction of the loops and turns reflected not the placement of obstacles
but variations in error signals relating to things other than obstacles.
It's perfectly possible that the ant is just as "complex" (another
weasel-word) as its environment, and that depending on circumstances,
anywhere between 10% and 90% of the loops and turns are due to variations
in the ant's reference signals. Simon's "need not" and "might" suddenly,
without reason, turned into "is not" and "is". Why did that happen, when it
was equally possible that the correct answers could be "is" and "is not"?
something else must have come into play -- perhaps a conviction that ants
ought not to be complex, and that the environment ought to be responsible
for most behavior.



>Exactly what those "rules" (system equations) are is not the subject of this
>discussion, no matter how much you insist that they are.
>
>>Of course you are implying that if there were obstacles, the ant would not
>>travel in a straight line toward its goal. That doesn't follow. If you see
>>an ant traveling in a straight line, chances are that it's counteracting
>>disturbances too small for you to see without a magnifying glass, thus
>>leading you to think that there aren't any. Or you may be assuming that
>>because the path remains undisturbed, there must not be any disturbing
>>forces acting. It may be puzzling how one ant could climb right over the
>>back of another ant while both ants continue on their paths, but you can
>>always shrug and say that the net force must have just happened to come out
>>zero for both ants.
>
>A masterly example of a straw-man argument.  You have said nothing here that
>has anything even remotely to do with Simon's conjecture.
>
>>Your observations are too limited. An ant often deviates abruptly from its
>>path when nothing at all is in its way. During our campground observations
>>of individual ants, Tom and I would often look at each other and say "Now,
>>why did it do THAT?" I doubt that you could find any correlation between
>>the placements of small obstacles and the path of the ant, especially if
>>you plotted the path from above.
>
>More of the same.  I can't tell if you're _missing_ the point or
>deliberately trying to obfuscate it.
>
>>It's a simple point, Bill, and one with which you of
>>>all people should have no quarrel.  The complex waveform of the plot of
>>>handle movement in a tracking task mirrors the complexity of the
>>>distrubances acting on cursor position, not any complexity in the five-line
>>>Pascal model of the control system that can accurately simulate this
>>>behavior.  HPCT is fundamentally simple, yet it is capable of generating
>>>extremely complex actions if the environment requires them.  Not so?

>>Yes, I understand your meaning here, but it is not Simon's meaning. I
>>repeat, Simon considered the organism to be one gigantic input-output
>>function, and said so in so many words.
>
>Ah, now I understand the high-gain resistance.  I really don't have any
>evidence that Simon held this view; it certainly was not expressed in
>_Sciences of the Artificial_, at least as I recall it now.  And it isn't the
>view I am attempting to defend here.  How did we shift from a simple
>conjecture Simon made about the origins of the complexity of observed
>behavior to a discussion about open- versus closed-loop models?  Simon's
>conjecture does not entail an open-loop model, even if he had one in mind
>when he made it.

OK, who was it who quoted Simon on the net recently about the input-output
function? He said it more than once. He believed it. And he was simply
reflecting the Zeitgeist, to the effect that the environmment is the
ultimate determinant of behavior. This was Simon's way of showing his peers
that he had not strayed into apostasy -- that he was still among the
annointed.

Simon's conjecture entails SOME model whether he knew it or not. As I
pointed out, he assumed that an ant saw the world the same way he did, and
had purposes like his. His "description" was loaded with private
interpretations.

Bruce, I have a right to reject Simon as a major player in this game. I
lived through the same years he lived through and worked on the same
problems using the same knowledge. But I did it the hard way: not with airy
"conjectures" but by working out models that would actually behave in the
right way. I did it without fanfare, and I kept my mouth shut until I had
something concrete to show. Simon went for the spotlight, and stayed in it,
so people got used to considering his off-the-cuff remarks as containing
wisdom. After all, if someone is well-known, he must know a lot about
everything, right? Thus we have a country in which someone can capture
attention by saying "I'm not a doctor, but I play one on TV." That "but"
carried loads of subtle implications, the main one being that playing a
doctor is the next best thing to being one.

Best,

Bill P.