the Test

CSGnet Archive CSG1995
1

[From Bill Powers (951103.1620 MST)]

Martin Taylor 951103 13:10 --

     As the trials go on, the reorganizations always succeed when a side
     effect of some random behaviour has the right result. ... The fact
     that the cat gets out quicker and quicker on successive trials has
     no bearing on whether the stick appears in any of the cat's
     controlled perceptions. ... If what I said above is true, then what
     the cat is "doing" has no connection with the door opening, except
     by coincidence unknowable to the cat. There are myriads of other
     things that coincide with "marking the territory and lying on the
     left facing the clock." So if you introduce disturbances, you will
     be testing for controlled variables that will change unpredictably
     if "mttalotlftc" no longer involves brushing the stick. It will
     appear to you as if those variables had not been controlled. How
     then could one study what the cat is "doing" in escaping from the
     box?

In doing the test to see what the cat is controlling as a means of
getting out of the cage, I would start with the most obvious guess: it
is seen brushing against the stick, which suggests control of tactile
sensations. To test that idea, I would move the stick slightly, just
enough to require a modification of the cat's actions to maintain the
brushing. If the cat was actually controlling "mttalotlftc", the cat
would compensate by changing its position during the "marking" phase,
assuming that marking requires contact. However, I probably wouldn't
characterize the behavior as "marking," because that's an interpretation
by the observer that goes beyond the observation. I would just say that
that cat is controlling for the sensation of rubbing against the stick.
I might check this out by spraying a topical anaesthetic on the area
that is being rubbed, and see if the cat moves to bring another area in
contact, or increases its rubbing pressure to bring the level of the
sensation higher.

     If what I said above is true, then what the cat is "doing" has no
     connection with the door opening, except by coincidence unknowable
     to the cat. There are myriads of other things that coincide with
     "marking the territory and lying on the left facing the clock." So
     if you introduce disturbances, you will be testing for controlled
     variables that will change unpredictably if "mttal- otlftc" no
     longer involves brushing the stick. It will appear to you as if
     those variables had not been controlled. How then could one study
     what the cat is "doing" in escaping from the box?

All effects of instrumental behavior are unexplainable at first, and
maybe forever, depending on the intellect involved. Most people, when
they flip a switch to turn a light on, have no idea why this arbitrary
change of position of a little lever on the wall makes the light bulb
light up. They don't need to understand electricity to learn this. Most
of what we know about the world consists of knowing that if we control
this perception in a certain way, that one will change in a certain way.

     So if you introduce disturbances, you will be testing for
     controlled variables that will change unpredictably if "mttal-
     otlftc" no longer involves brushing the stick. It will appear to
     you as if those variables had not been controlled. How then could
     one study what the cat is "doing" in escaping from the box?

When you introduce disturbances, you aren't trying to keep control from
succeeding. You're just putting in a little test perturbation to see if
it's resisted. There seems to be a temptation, however, such that when a
little disturbance fails to have an effect, you want to keep increasing
the disturbance until it _does_ have an effect. That's not the principle
of the Test. The principle is to apply just enough disturbance to the
variable under test that if it is under control, it will vary less than
you would predict (and perhaps not measurably).

I'm not sure what you have in mind when you say

     ... if "mttal-otlftc" no longer involves brushing the stick.

Since "marking the territory and lying on the left facing the clock" is
what the cat does to escape, how would this no longer involve brushing
the stick? Brushing the stick is what opens the box. If the cat stops
marking the stick, the box will no longer open. Our disturbances won't
change that. Please explain.

     When the environment is relatively unchanging, the side effects of
     our actions are also relatively stable, and we may achieve our ends
     by means that are radically different from the way we think we are
     achieving them.

You're talking about the environmental feedback function: the link
between our actions and their effects on a controlled variable. We never
need to know the nature of that link.

     If, to take an example I assure you to be hypothetical, I
     habitually bring my wife flowers on Fridays, and she expresses
     pleasure and acts generally as I would wish, I may think she likes
     flowers, and try bringing more if I think she is in a bad mood.
     But maybe she just enjoys the silly smile I have when I hand her
     the flowers, and a toy train might serve just as well.

Bringing her flowers entails more actions than you may be aware of, and
the actual link between those actions and the results you experience as
a consequence may be totally unknown to you. This is irrelevant to the
Test; I could still determine (by conspiring with her) that you want
your wife to express pleasure etc., and that your means of doing this is
to bring her flowers (and all that regularly goes on when you do so). I
could discover that you are controlling for bringing flowers to her by
creating difficulties in obtaining them, suggesting taking her a toy
train, etc.. Even if the toy train would in fact give you the same
results, you don't know that, so you are controlling the variable you
perceive as being linked to the result. Whether it is actually linked to
the result in the way you think is irrelevant.

     Anyway, I hope you see my puzzlement: when you set up an experiment
     in a fixed environment, you cannot be sure whether the actions you
     observe are behaviour or side-effect, but when you change the
     environment you may change the behaviour that you are trying to
     observe. Only if you happen to hit on a perception whose control
     is unaffected by the change of environment will you even begin to
     be able to apply the Test.

But isn't changing the environment a disturbance?

···

-----------------------------------------------------------------------
     This is what I was taught to do as a grad student in psychology.
     It was then called "converging operations," following (I think)
     P.W.Bridgeman. My professor (Tex Garner) thought that it was the
     only legitimate way to investigate nature. So I don't understand
     why you follow up with:

     >I far as I have ever been able to see, psychology is simply not
     >conducted in this way.

How many psychologists actually do what Bridgeman recommended? When a
psychologist publishes a report that some stimulus or situation causes a
particular change in behavior, where are the converging operations that
test the causal assertion?

Incidentally, I have received your paper, and think it is about the best
job that could have been done with those experimental results (even
without any converging operations).
-----------------------------------------------------------------------
Kent McClelland (direct post) --

Very mysterious. I'll check it out.
-----------------------------------------------------------------------
Best to all,

Bill P.

2

[Martin Taylor 951103 19:30]

This will be brief, as I am about to go home...

Bill Powers (951103.1620 MST)

I think your answer is at cross-purposes to my question.

                                       There are myriads of other
    things that coincide with "marking the territory and lying on the
    left facing the clock." So if you introduce disturbances, you will
    be testing for controlled variables that will change unpredictably
    if "mttalotlftc" no longer involves brushing the stick. It will
    appear to you as if those variables had not been controlled. How
    then could one study what the cat is "doing" in escaping from the
    box?

In doing the test to see what the cat is controlling as a means of
getting out of the cage, I would start with the most obvious guess: it
is seen brushing against the stick, which suggests control of tactile
sensations. To test that idea, I would move the stick slightly, just
enough to require a modification of the cat's actions to maintain the
brushing. If the cat was actually controlling "mttalotlftc", the cat
would compensate by changing its position during the "marking" phase,
assuming that marking requires contact.

That last "assuming" is where the cross-purposes start. I'm assuming that
marking does NOT require contact, but the box configuration along with
whatever else the cat perceives just happens in this case to generate
contact as a side-effect. The notion of its being a side effect was the
core of my posting.

      However, I probably wouldn't
characterize the behavior as "marking," because that's an interpretation
by the observer that goes beyond the observation. I would just say that
that cat is controlling for the sensation of rubbing against the stick.

And I assumed that the cat had no perception of rubbing against the stick,
and was controlling for completely different things.

I'm not sure what you have in mind when you say

    ... if "mttal-otlftc" no longer involves brushing the stick.

Since "marking the territory and lying on the left facing the clock" is
what the cat does to escape, how would this no longer involve brushing
the stick? Brushing the stick is what opens the box. If the cat stops
marking the stick, the box will no longer open. Our disturbances won't
change that. Please explain.

That's the problem I was trying to ask about. The cat has to do something
else, such as miaow while looking at the picture on the East wall, which
happens to brush the stick, again as a side effect. The something else
is just as irrelevant to the stick as was "mttalotlftc" (I used that
nonsense acronym in part to suggest the "nonsensical" nature of the
successful action).

All that your test would do, so far as I can see, is to determine whether
the cat's perception of brushing the stick is controlled as part of the
escape procedure. I'm asking about the assumption that it isn't, based
on the notion that the side-effects of similar control have similar results
in a constant environment. That is, I think, the basis of how your version
of reorganization arranges for perceptual control to maintain the intrinsic
variables near their reference levels. The controlled perceptions have
nothing causal to do with the intrinsic variables, but in an environment
with reasonably stable feedback functions, the side-effects of controlling
the (correctly reorganized) perceptions is the control of the intrinsic
variables. So, I ask, may it not be with the cat?

Rather than following with more comments on your comments, I wonder if you
would care to re-analyze the situation with the above in mind. I'm talking
about reorganizations that have the effect of achieving the cat's reference
perceptions without the cat actually controlling for any perception that
includes the truly effective one as a component.

Most people, when
they flip a switch to turn a light on, have no idea why this arbitrary
change of position of a little lever on the wall makes the light bulb
light up.

In your example of turning the light switch, it really IS the light switch
that matters. To make a rather far-fetched analogy to what I am getting at,
suppose that you had learned that you can get light to happen by dancing
three steps when you come in the door, and flinging your jacket with a
swoop over your head. When the jacket falls, again, the light is on
(the upward sweep catches the light switch, but you don't know this,
because you've never noted the funny protrusion on the wall that other
people might call a "light switch"). And then you find that if you turn
your toes outward during the dance steps, the light doesn't go on. WE know
that it's because your jacket missed the switch on its upward sweep, but
YOU don't know that. To you, it's obvious that you have to be careful to keep
your toes pointing in during the dance steps. THAT's what makes light.

And in another room, the ritual is different, but you have discovered how
to make light there, too.

Now cast yourself in the role of experimenter, looking to see how a human
learns to make light.

···

---------------

How many psychologists actually do what Bridgeman recommended? When a
psychologist publishes a report that some stimulus or situation causes a
particular change in behavior, where are the converging operations that
test the causal assertion?

Usually it is another psychologist who sees the situation in a different
way an performs the converging operation: "If what Smith says is right,
then X should happen in situation Y that he never thought of."

I think what tends to be lacking is the same as what is lacking when
a newspaper prints an erratum. It doesn't get to the same audience,
and doesn't correct the original impression. Add to that the likelihood
that a negative result (does not) get published, and maybe your original
comment is somewhat justified. I think it's a sociological problem, though,
not a psychological one--I mean the sociology of the science, not the
science of sociology:-)
--------------
Good night (I typed "Goof" before I edited it--perhaps I should have left
it that way). It's a lot longer than I had anticipated when I began.

Martin

3

[From Bill Powers (951107.1620 MST)]

     You are still taking the experimenter's view rather than the cat's.

Basically, I'm having trouble figuring out what point of view we're
supposed to be taking here.

     I don't know whether an experimenter could determine what the cat
     is controlling during any specific escape sequence. I don't care.
     The question is about what the cat might be doing, not the
     experimenter. I know I started by asking about the application of
     the Test, but the Test leads into the organism tested, doesn't it?

From the cat's point of view, I suppose, what is done is to control

different perceptual variables in different ways until the door opens.
When the cat stops reorganizing it will keep controlling the same
variable the same way until the door fails to open; then it will
reorganize some more until the door opens again.

It should always be possible to discover what the cat is actually
controlling by applying disturbances and seeing if they are resisted,
and of course then going on to complete the Test.

        During rapid reorganization?

Well, no, of course not. There would be nothing regular to observe.
Wouldn't we, the experimenters, be able to realize that? We might search
for a while for some regularity, but eventually we'd catch on that the
cat is still reorganizing the rules and isn't actually doing anything
the same way from one trial to the next. In fact, we would probably
conclude that that catc can't find a solution in the form of a regular
control system, and has fallen back on the E. coli method: trial and
error until the door opens, every time.

     By hypothesis throught this interchange of messages, the cat
     doesn't perceive the contact. So, again by hypothesis, preventing
     it from sensing contact will not destroy control--the cat escapes
     nevertheless.

So you're saying that there is actually no first-level control system.
The cat is escaping every time just by producing random movements until
the door opens.

     By analogy, in the world in which we live--society and all--if we
     do certain things, other things happen (usually). We need never
     know why, so long as God doesn't disturb the environment as you
     propose your experimenter to do. So long as the feedback paths
     remain unchanged, if we discover a way to influence our
     perceptions, we can control them. It is inconsequential that some
     element of the feedback path is a pure side-effect of controlling
     another perception for which the main output sets the reference
     levels, so long as the environmental feedback paths don't change.
     But when they do, those side-effects may be quite different.

Sniffing carefully for traps, I can't find anything in this to disagree
with.

     So with your disturbances to the cat. When you change the cat's
     environment, it may well go back to "random" behaviour, changing
     its actions until some different side-effect of some different
     perceptual control system happens to have the right influence on
     the main controlled perception--being out of the box and at the
     food.

Why should the cat go back to random behavior if it is still controlling
successfully? You seem to equate "disturbance" with "disruption of
control." A disturbance is just a mild perturbation, no larger than is
required to see that the control relationships exist. It doesn't
necessarily change the controlled variable by a perceptible (to the
experimenter) amount. If you're driving and I guess that you're
controlling the lateral position of the car, I can apply a disturbance
to the steering wheel and see that you resist it, and do so quite safely
as long as I make sure you still have control.

In
other words, you are saying that there can be a reliable coincidence
that will fool us. Considering how unlikely such coincidences are, is
this really a serious problem?

     In view of the above, do you think that it is either unlikely or a
     problem?

I guess I just don't see what you're getting at. What I thought I was
talking about was a case where I apply a disturbance and a random change
in the cat's behavior just happens to oppose the effects of the
disturbance, without the cat's actually sensing and controlling the
variable to which I'm applying the disturbance. I say that this sort of
coincidence is unlikely because we don't just apply one disturbance and
then sit down to figure out what the controlled variable must have been.
We apply every kind and amount of disturbance we can think of, again and
again, to make sure we have not omitted any dimension of the controlled
variable and that the identification is secure. The coincidence of which
I speak would have to involve a random behavior which, again and again,
supplied just the right amount and direction of opposing force to cancel
the effects of my disturbances. I may be missing something, but I would
consider that sort of coincidence astonishing, not something to worry
about very much.

     I presume it is through attaining control and retaining it as the
     environmental feedback functions change that the cat (and we) learn
     what is really the effective perception to control in order to
     control the higher-level one. Once the feedback path ceases to
     depend on side-effects that change with the environment, it becomes
     reliable and control consistent.

The changes in the environmental feedback function are not part of the
equation. What we learn is that controlling A in a certain way (which
can be done over quite a range of environmental feedback functions and
additive disturbances) will reliably produce B. We never need to know
WHY that is so. We never need to know the nature of the environmental
feedback function relating our action to A, or the nature of the actual
effect A has on B. In particular, we never need to know what is "really
the effective perception to control" as opposed to "the apparently
effective perception to control." All we need to know is what is
sufficient, not what is necessary.

If we want to know what is necessary, a lot more work is involved. But I
presume that is not the cat's problem.

     I hope you see now that it is precisely the variation in the
     "effective side-effects" that is at the heart of what interests me
     in this situation.

I can't say that I do see what you're getting at. Every now and then I
think I do, and then you say something that makes me think I don't. I
guess I don't.

···

---------------------------------------------------------------------
Best,

Bill P.

4

[From Bill Powers (951109.0940 MST)]

Hans Blom, 951109 --

It would help if you were to take a more creative position on some of
these questions instead of just trying to zing fast balls past me. I
don't mind the glove practice, but I'm the only one getting the
practice, whereas I'm not the main one who needs it.

     I had and responded to a different understanding of what The Test
     comprises: to identify _which_ variable is under control, i.e. to
     test a number of hypotheses, in parallel (often not possible) or
     sequentially (which may take a lot of time). Your remark clearly
     applies to testing _one_ hypothesis only. That is not the full
     version of The Test. One can determine whether _one particular_
     hypothesis is correct. If not, it appears to me, the chance for
     later testing of other hypotheses _in the same situation_ is
     irredeemably lost, at least in the same subject. Assuming and
     testing _one_ controlled variable at a time will work fine. But is
     that The Test?

Why not answer your own implied questions? How would you go about
applying the Test to more than one proposed variable at a time? I have
no problem with thinking of ways, but if I just tell you, what will that
accomplish? The Test involves testing for _which_ variable as well as
_how much of a particular_ variable that one hypothesizes to be under
control. If you have any serious problems with this, I'll be glad to
tackle them, but since I already know an answer to all the simple stuff,
I wouldn't get much out of it. The Test isn't a recipe that you have to
follow blindly; it's just an organized way to find out whether a
particular behavior fits the PCT model. If you understand the PCT model
you can think up your own tests; they probably won't be much different
from mine. Really, can't you think of any way of applying the Test to
multiple variables at the same time?

How would you test to see if a particular example of behavior involved
model-based control as opposed to traditional control or S-R behavior?
Wouldn't you go through similar procedures?

···

--------------------------
     "Accusation"? Is that how I come across?

Well, when you talk about the "Achilles Heel" of PCT, you do come across
as a bit aggressive.
--------------------------
     By the way, I do not believe in tests that are designed to give
     yes/no answers.

What makes you think I do? The Test can reject a bad guess, but there's
grey area within which you have to decide how much effect a disturbance
can have on a variable that still permits the variable to be called
"controlled." I've said this many times, and so have others. Do I have
to say it every time I mention the Test?
-----------------------------------------------------------------------
Best,

Bill P.

5

[Martin Taylor 951122 10:45]

Bill Powers (951121.2100 MST)

Only a short comment this time (I hope)...

The first step of the test, the main one on which you're concentrating,
is primarily a method of ruling out a false identification. If a
disturbance is applied and it has exactly the effect predicted under the
assumption of NO control, you can stop because you have shown that the
postulated variable is not being controlled. In all your examples, you
are being vague about just what the postulated variable is, the exact
kind of disturbance you would apply, and just how you would predict what
the effect of the disturbance would be if there is no control. And I
think you are interpreting the test to mean that if there is apparent
resistance to the disturbance, the Test has proven that control is
taking place, which is not the case. All it has then shown is that
control is not ruled out, a very different matter.

If you have been unable to rule out control using one disturbance, there
is nothing to prevent you from trying to rule it out by using other
disturbances.

There's no need to be other than vague, I think, because the situation is
such that no matter what the disturbance, the cat appears to resist it
accurately (if possibly slowly). The cat, if it escapes at all, always
moves the stick in such a way as to negate the effect of any disturbance
you might have introduced.

If I may requote from Chuck Tucker's quote from you:

       9. If all of the above steps are passed, you have found the
          input quantity, the variable the person is controlling (118).

All the above steps ARE passed, and for all (suitably small) disturbances
you choose to apply that affect the cat's movement of the stick, and for
all modifications to the cat's ability to sense that do not alter its
ability to control those perceptions that result in its moving around
the cage.

I take it that what Chuck quoted is therefore not a correct description of
the Test as you now conceive it.

I'm interested in what the experimenter can discover about the cat, but
I'm more interested in what the cat could possibly discover about its own
world--what the cat can learn to control that involves the (known to the
experimenter) real escape mechanism. This latter seems to me to relate
very much to the differences between science and religion, and to one
of Hans Blom's comments that I paraphrase as putting together the ghosts
of many different successes.

Martin