loopholes in the Test

CSGnet Archive CSG1995
1

[From Bill Powers (951122.1100 MST)]

Martin Taylor 951122 10:45 --

     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.

Well, let's be a little more specific anyway. You say, the cat always
moves the stick in such a way as to negate the effect of _any
disturbance_ you might have introduced. But here you seem to be ignoring
the fact that the disturbance is always applied to a specific variable
and not to others. What you are always testing for is resistance to the
disturbance of the specific variable that is being disturbed -- not to
any side-effects of that variable. If I apply a disturbance to the stick
and the cat escapes anyway, the escape has nothing to do with the test;
I'm applying the disturbance in a way that should alter the stick
pressure, and if it does, the stick pressure is not under control _even
if the cat escapes_. If I were testing for control of escaping, I would
put a barrier up in the opening, put a spring behind the door, and do
other things that the cat would have to overcome to exit from the box.

You seem to be postulating a situation in which there is no way to
separate the movements of the stick from changes in other variables that
can affect the stick. I really can't imagine that being literally the
case. It should be possible even to separate the cat's controlling for
pressure on the pad of a foot held against the rod from the movements of
the rod itself.

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

Well, in one sense it is correct: if the person applies the test with an
understanding of its point, nothing needs to be added. However, if the
person is simply running down the list and checking off steps, like
making a cake from a recipe, a lot more detail would have to be added,
in which every contingency is handled. The difference is in whether you
see the test from the principle level or the sequence level. If you see
the principle, you can make up your own programs and generate different
sequences in every specific case -- for example, sometimes you might
observe right away that there's no way for the person to perceive the
variable you've picked as a candidate controlled variable, so you can
apply that step first and skip the others. Basically, all you're doing
is looking with your mind's eye at the canonical control-system diagram
and checking to find counterparts of all its visible functions in the
behavior you're testing.

I suppose, in the light of what you've been saying, I could add a
general instruction: Keep repeating every step that is passed until
you're sure you haven't missed a way in which the test could be failed.

     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.

If the cat escapes, whatever it's controlling must have an effect on the
real escape mechanism. But why should a cat ever go farther than that? A
human being might wonder _why_ controlling one variable has a wanted
effect on another, but would a cat be curious about that? Lacking
symbol-manipulation (I assume), the cat wouldn't think about the problem
in the way we would. Would it? Learning to control A in order to control
B is, I should think, enough of an intellectual achievement for a cat.
It's enough for human beings, too, in most situations. How many people
understand why flipping a switch makes the light come on, or why turning
the ignition key makes the car start? How many care?

···

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

Bill P.

2

[Martin Taylor 951122 14:00]

Bill Powers (951122.1100 MST)

I don't know that your new subject heading fairly represents what I am
getting at, but let it ride...

Let's deal with what I see as a strictly analogous situation, but a simpler
one than the cat getting out of the box--a tracking task with a joystick
handle.

If I apply a disturbance to the stick
and the cat escapes anyway, the escape has nothing to do with the test;
I'm applying the disturbance in a way that should alter the stick
pressure, and if it does, the stick pressure is not under control _even
if the cat escapes_.

can be rephrased in context of the analogue:

--If I apply a disturbance to the stick
--and the subject continues to track well, the tracking has nothing to do
--with the test; I'm applying the disturbance in a way that should alter
--the force required for the subject to move the stick, and if it does,
--the stick pressure is not under control _even if the tracking continues
--to be precise_.

In this context, I don't think you would argue that making the feel of the
joystick heavier, requiring the subject to use more force in order to keep
tracking, would demonstrate that the force applied by the subject was not
a controlled perception. We assume that it is, and that a varying reference
value is the reason why the applied force changed when you altered the
feedback function of the tracking control loop.

You seem to be postulating a situation in which there is no way to
separate the movements of the stick from changes in other variables that
can affect the stick.

I don't think I am, at all. In my mind's eye I am seeing all sorts of
decouplings. Some of them affect the cat's escaping, and lead to rapid
reorganization, some of them don't. If they don't, they may well lead
to changes in the values at which the cat produces different physical
effects (such as pressure on the stick), but the values the cat produces
wind up being adequate for continuing control of the variable that matters--
perceiving itself to be out of the box. So with changing the required
force, the length, the position...of the joystick in a tracking task,
in which we presume the relevant perceptual variables ARE being controlled.

Now, in the tracking experiment analogue, if you provided a joystick and
a mouse and a touch-pad, but connected only one of those up at random for
any one tracking run, you would find out that pressure on the joystick
was not really a critical controlled variable. Once the subject got
over the reorganization phase of finding that now it is the mouse that
affects the cursor, not the joystick, new controlled intermediate variables
would come into play.

In the tracking experiment, we assume that the intermediate variables
required by the physics and electronics really are the CEVs corresponding
to controlled perceptions. In the cat experiment, we are using as a
base condition that the cat affects those physical variables only as
side effects of things it is really controlling. Can we tell the
difference? Can the cat-subject?

But why should a cat ever go farther than that? A
human being might wonder _why_ controlling one variable has a wanted
effect on another, but would a cat be curious about that? Lacking
symbol-manipulation (I assume), the cat wouldn't think about the problem
in the way we would.

I haven't introduced any question of "why", but if you want to, I must say
that I agree with you.

How many people
understand why flipping a switch makes the light come on,

If you remember one of my fanciful examples, I put a human in the same
position as the cat. Knowing nothing of light switches, the human discovered
that if he entered a room with three dance steps and then flipped his jacket
over his head, the light came on. But it didn't if the dance steps were
performed toes-out. The human might wonder about the "why" of why toes-out
offended the gods of light-giving, but that wouldn't affect his failure to
discover "what" made the light go on, which we omniscients know to be the
catching of the jacket on the light switch.

As experimenters, we might be amused to see that our human-subject had
found a rather exotic way to turn on the light, but we would observe that
he consistently did turn it on. Does he purposefully flip his jacket? Sure
he does. That's perceptually controlled. Does he purposefully flip the
light switch? No (we omniscients know), because he has not noticed that
it exists. How, as non-omniscient experimenters, do we discover whether
he purposefully flips the light switch, if that's the ONLY way the light
can be made to go on?

Martin